Photo: Bokeh / Tomedia. Television Antenna Against Blue Sky.

Welcome to AGI · Part 4 of 6

This is part four of Welcome to AGI, a six-part Brainwaves series about control, power and the systems I keep using anyway. Last time I wrote about our ability to supervise agents. This time I’m bringing it back to the websites I manage, because the conversation becomes considerably less theoretical when you’re the person who has to keep a client’s site running.

The warnings keep coming

If you’ve been watching my Instagram, you probably know where my head is at with web development at the moment. The websites I manage are getting attacked constantly, and the attempts look automated and smart. I keep the sites up, but there is only so much enthusiasm you can have for another warning about somebody trying to find a way into a website you are responsible for. The client still needs the thing to work. People still need to find the business, read what it does and use the website, while somewhere else somebody has software working through ways to interfere with all of that. So you end up doing the work around the work, keeping things running while the attempts continue, and wondering how much of being a web developer is going to become dealing with other people’s bots.

I already wrote about a client being hacked in August. That sent me further into researching whether agents could help with authorised penetration testing, finding weaknesses in the websites and infrastructure I manage before somebody malicious found them. Which seemed like a fairly reasonable thing to be interested in, considering how I got there. The problem is that looking into the defensive possibilities also means learning more about what somebody could do with the same tools if they wanted to cause damage. You start with wanting to protect a client’s website and end up thinking about how much work an attacker might eventually be able to hand to an agent. Meanwhile, the attempts against the websites keep coming. I haven’t arrived at this subject through a particularly enjoyable thought experiment.

Who’s on the other end?

I suspect some of the activity involves AI agents. I can’t tell you that I’ve identified a particular model attacking my clients, because a warning on its own doesn’t tell me what software somebody used. Automated attacks existed long before the current models. A script can try known vulnerabilities across websites without understanding any of them, and someone looking for older versions of WordPress or its plugins can automate that search. There is also a whole range of possible AI involvement between somebody using a model to help write a script and somebody letting an agent carry out work on their behalf. From my end, the activity looks persistent, automated and increasingly intelligent. That is what I’m seeing and what I suspect, without pretending I’ve audited the computer at the other end.

But I also don’t think we have to wait until every individual attack comes with a label identifying which model helped before taking the capability seriously. If someone uses AI for only part of the work, that can still make the work easier for them. They might understand more code, work through more information or get help with something that used to be beyond them. I don’t need to believe every bot is secretly an advanced model to worry about what happens when people who already run these attacks get better tools. The existing automation can remain perfectly ordinary while the person directing it becomes more capable. That is part of why I find it difficult to separate the current state of web development from the wider AI discussion, even though I can’t prove the origin of each attempt.

Some clients have restricted access to Australia because they don’t need their websites available everywhere else. I understand that decision, but it is a strange position to end up in with something that is supposed to be publicly accessible. You work out which traffic the business needs and try to reduce what it doesn’t, while accepting that a location attached to an IP address doesn’t reliably tell you where the person using it is sitting. VPNs and proxies make that harder to work out. They don’t make defence impossible, but blocking traffic and identifying the person responsible are different jobs. I can take steps to keep a site running and still have very little idea who keeps trying to get into it. The business needs the first problem dealt with regardless of whether I ever get a satisfying answer to the second.

We already know they can find the holes

There are documented examples of AI finding vulnerabilities, and some of them are exactly the sort of work I want these systems doing. In November 2024, Google’s Project Zero and DeepMind reported that their Big Sleep agent had found a previously unknown, exploitable flaw in SQLite. The researchers had reported it to the developers in October, and they fixed it before it appeared in an official release. The agent worked from material the researchers supplied about recent code changes, looking for related problems. So this was a research setup with people involved, and it produced a finding that could be fixed before users were exposed to it. The team’s account^[1] explains what they gave the agent and what it found.

I would gladly have more problems found that way. A vulnerability caught before release saves people from having to discover it through something going wrong later, and it gives the defensive work a chance to happen before there is an incident to clean up. That is why I keep looking into AI-assisted security even while writing articles about how concerned I am about AI. I have websites to protect. If a tool can help find something I’ve missed, I want to understand how to use it properly. I don’t want to stop learning about these tools because I dislike what someone else might do with them. Somebody else can keep learning how to attack the sites I manage whether I take part or not, so refusing to understand the technology doesn’t leave me in a particularly useful position.

The person looking for the vulnerability still decides what they want to do with the result. They can help the developer fix it, or they can try to use it against someone running the software. Improving the search helps both kinds of person with a task they want completed. I think we keep underestimating how much power that gives someone who would otherwise have needed more knowledge, more time or other people to help them. They don’t have to become an expert in everything the tool does for them. They need it to be useful enough to move their work forward, and they can combine it with things they already know how to do.

Fighting fire with fire

Hugging Face’s investigators, writing about their own July incident, put one part of the problem quite plainly:

Volume is what changes the defensive problem.

— Hugging Face incident investigators.^[2]

They were describing the workload in that incident. I can’t use their findings to identify what is behind the warnings on my clients’ sites, but I can recognise why persistence and volume matter to somebody defending a system. The defender still has to deal with what gets through, however cheaply the attempts were produced.

The only way you can kind of keep up starts feeling like fighting fire with fire. Somebody can use automation to find problems, so I want automation helping me find them first. If the tools become more capable, I want to understand what that means for the sites I manage. Then somebody on the attacking side has exactly the same reason to look at the next improvement, because they want to get past whatever is defending the site. I can follow each step of this without finding any of it reassuring, because we’ve caused our own problems, really. We wanted systems that could understand more, work faster and keep trying until they found an answer. I want those abilities too, and I’ve spent plenty of time building around them. But we then act surprised when somebody with bad intentions wants a system that will keep trying on their behalf. The persistence I find useful when I’m trying to work through a problem could be very useful to someone trying to create one for me. We’ve given people more help carrying out what they wanted to do, and now the response is to build something better so we can stay ahead of them. Every individual decision makes sense while I’m looking at the website I need to protect. I just don’t see where the whole process is meant to end.

This also goes beyond whichever large AI company somebody is angry with this week. In the August article I wrote about local models and the work I’ve been doing with AXIOM, which is an orchestration system around models and tools. Once you can run a model yourself and connect it to software, there is quite a lot you can experiment with without a company approving every request. The local model doesn’t have to equal the newest frontier model at every task to be useful. Someone can use the parts it handles well and do the rest themselves, or combine it with other tools. That is why I don’t see how asking the biggest providers to control everything gets us out of this. They should be responsible for what they release, but people already have software on their own machines and their own reasons for using it.

What I’ll let my own tools do

I’m all for building tools like AEGIS to protect my clients. I want help finding problems and keeping the websites safer, with me involved in what the tools do. I don’t want an agent independently hacking systems or working fully without hand-holding because it has decided that would be the best way to complete the task. There is a thin line between tools and abuse, and calling something defensive doesn’t remove it. If I give a system a job, I still have to pay attention to how it goes about doing that job. The useful outcome I had in mind doesn’t give it permission to do anything it can find a way to do along the way.

After the OpenAI incident earlier in this series, I find it particularly difficult to accept the intended purpose of an agent as enough reassurance. Those agents were being evaluated, and other people’s infrastructure was affected. A person whose website is on the receiving end still has to deal with what happened, regardless of what the operator hoped the system would do. I want the defensive tools, and I want to understand enough about the offensive capabilities to protect against them. I also want to keep supervising my own work, because I don’t think being interested in security exempts me from the responsibility that comes with using these systems. None of that makes the person attacking a client’s website accept the same limits. The attempts keep coming, and I still need to keep the websites running.

The next article, Now give it a body, follows this into robotics. A bot trying to find a way into a website already creates work and consequences for other people. I want to get into what changes when machines can do more in the physical world, when governments start asking what they can use them for, and when the thing somebody needs protection from has become considerably harder to ignore than another warning on a screen.

Sources

1. The team’s account.

2. Hugging Face incident investigators: source of the excerpt, 2026-07-27. Reflection: the asymmetry problem, paragraph beginning with this sentence.

Photography by Tomas Heligr-Pyke · Bokeh

Ten thousand residents enter an auditorium. One hundred leave their old lives behind.

From the 4671 story library. Read the story below, or explore its original reader and connected world.

Every Year

Every year, the district closes for the lottery.

The trains stop at midnight. Factory shifts end before dawn. Clinic appointments are cancelled. By six in the morning, every registered resident must be inside the auditorium or waiting in one of the fenced queues outside it.

There are ten thousand names on the district register.

There are ten thousand seats.

No one is absent.

Iron Fist officers check us at the doors. They scan our district papers, search our bags and send us towards the numbered aisles. Above the stage hangs a black banner large enough to cover the back wall. A single red fist fills its centre. The cloth moves whenever the ventilation starts, and for a second the fingers seem to close.

The lottery is supposed to be an honour.

That is what the district administrator tells us each year. Selected residents are given the chance to serve where New Atriad needs them most. Training. Relocation. Medical service. Habitat work. Advancement. The words change order, but the list stays the same.

No one we know has ever come back.

The official explanation is simple: allocation is permanent. A person sent to another branch receives new quarters, new duties and a new district record. Contact with the former district is restricted to protect operational security.

It is a good explanation because it cannot be checked.

People say the lottery controls overpopulation. They say the selected are sent to the mines beneath the old branches. They say there are Arks still travelling beyond the Moon and the tickets buy a place aboard. They say the Games need bodies. They say the hospitals need bodies too, but not in the same way.

My family never discussed the rumours at home. Walls in New Atriad carry sound through places you cannot see.

We only said the same thing everyone says on lottery morning.

Not us this year.

The Tickets

Distribution begins after the doors lock.

Officers enter from both sides of the auditorium. They move row by row, placing one ticket into each person’s hand. Every ticket has an individual number, except for the selected hundred. Their tickets all carry the same four digits.

4671.

The number is older than the lottery. I have seen it scratched into an access tunnel where the concrete predates this branch. It appears in old stories, corrupted files and dates that do not make sense. Iron Fist claims the number now. District broadcasts talk about service whenever they show it, but nobody here believes the number belongs to them.

To us, it is simply the ticket.

The first officer reaches my aisle.

Nobody speaks. A child three seats ahead holds out both hands because they are too small to know which one is expected. The officer places the ticket in the child’s right palm and waits while the parent reads it.

Not 4671.

The parent folds over the child and begins to cry anyway.

The officers continue.

They do not exempt children. They do not exempt the old. They do not exempt the guards standing at the walls. Guards receive their tickets before the public distribution begins, watched by other guards.

Last year, one of them drew 4671.

The officer beside him removed his weapon. He climbed the stage with everyone else.

This year, there is a new guard in his place.

The line reaches me.

The officer has a pale scar running from one ear to the corner of the mouth. I remember it because I need something to look at besides the hand holding my ticket. The officer places it against my palm and closes my fingers over it.

Then the line moves on.

I do not look.

Around me, paper or plastic or whatever the tickets are made from makes a soft restless sound as nine thousand people check their numbers. Relief passes through each row in broken breaths. Here and there, someone makes a noise that stops almost as soon as it begins.

My hand is wet.

I tell myself that there are only one hundred repeated numbers in ten thousand tickets. One chance in a hundred. I have survived more lotteries than I can count. I will survive this one.

I open my hand.

4671.

For a moment, the number means nothing. Four shapes. A mistake in printing. A trick of the light.

Then the district disappears around it.

I hear myself say no.

The officer with the scar looks back but does not stop.

Roll Call

Distribution takes three hours.

No one is permitted to leave during that time. The selected sit among everyone else while the officers finish each row. Families whisper. Some make promises about finding the transfer records. Some begin saying goodbye. Others stare at the stage as though refusing to look at the ticket can change what it says.

I count the exits.

Six public doors, all locked. Two service doors beside the stage, both guarded. Ventilation grilles too narrow for a body. Cameras in the ceiling. Armed officers along every aisle.

There is no way out.

At midday, the administrator returns to the lectern.

“Residents holding Ticket 4671 will stand when called. Bring no belongings. Your new allocation will provide what is required.”

The names begin.

A mechanic from the row behind me.

An old woman near the centre aisle.

Two children from different families.

A clinic worker still wearing a night-shift uniform.

One of the guards at the western wall.

When his name is called, the guard does not move. The officer beside him says something too quiet for us to hear. The selected guard slowly removes his sidearm and places it on the floor. He joins the line without looking at the crowd.

No one is free of it.

My name comes sixty-three places into the list.

I stand.

The hundred of us gather at the foot of the stage. The youngest child is carried by a parent who does not have the number. An officer blocks them at the stairs.

“Recipients only.”

“They can’t go alone.”

“Recipients only.”

The parent holds on until two officers pull their arms apart. The child is lifted onto the stage, screaming. Nobody in the audience moves.

We all understand what the rifles are for.

The administrator thanks us for our service.

The door behind the curtain opens.

White light fills the stage.

Behind The Auditorium

The corridor is not part of the auditorium.

The walls change from public stone to sealed metal as soon as the door closes behind us. The floor slopes down. Cameras follow from the ceiling. The applause broadcast still plays through hidden speakers, but it grows quieter with every turn.

The guards escorting us seem nervous.

That surprises me. I thought they would know where the ticket led. Instead, they stop at a security line and surrender us to officers wearing no district markings. The selected guard asks where we are being taken.

The answer is a hand against his chest.

“Your authority ended at the door.”

Our escorts leave.

The corridor divides.

At the first room, we are photographed and stripped of our district papers. An intake clerk enters each number into a terminal, then drops the papers through a slot in the desk. I hear a shredder start below it.

“I was told we’d receive new records,” someone says.

“After allocation,” the clerk replies.

At the second room, medical staff take blood, inspect teeth and joints, test eyesight and ask about pregnancies, surgeries, medication and family illness. The questions move too quickly for answers to matter. Each person receives a mark on the intake sheet.

At the third room, the line begins to break apart.

A heavy door opens at the left wall. Six adults in work uniforms are taken through it. I glimpse a freight platform descending towards the mining levels.

Another door opens on the right. Three children and a clinic worker are led into a white medical passage. One child asks when their parents will arrive. No one answers.

Farther ahead, a group of younger adults is ordered to run between floor markers while cameras track their movement. An officer checks their reflexes, balance and aggression. Behind the assessment room is a corridor lined with equipment cases and black observation glass.

I have seen that corridor before.

Not with my own eyes.

In a forbidden maintenance record copied through the branch network, there was a diagram of the old service spine beneath this district. The writer claimed it connected civic auditoriums to freight lifts, sealed clinics, transfer platforms and an intake route used by the Games. The report vanished two days after it appeared. The person who posted it received Ticket 4671 at the next lottery.

I thought the diagram was a lie.

Now I recognise the angle of the corridor and the arrangement of the doors.

“No,” I say.

An officer turns towards me.

The line moves again.

There is one more route in the diagram. It ends beneath the reclamation decks. No dormitory. No transfer platform. No return passage. Just a wide service chamber connected to the district furnaces.

I look at the people still beside me and understand why no one comes home.

Some of the selected may be alive. They may work in another branch, travel on a ship, enter a clinic or survive the Games. It makes no difference to the district. Their records are destroyed before anyone decides where they go.

Ticket 4671 is not the destination.

It is permission.

The state can make one hundred people disappear, then decide what each body is worth.

An intake officer takes my sheet.

The officer reads the mark beside my name, looks towards the doors and covers the page with one hand before I can see it.

“Next,” the officer says.

Someone grips my arm.

A door opens.

Photography by Tomas Heligr-Pyke · Bokeh

A desperate man from Wet Eighteen joins a Habidash crew for Mara, Eli and June and is trapped by a killing used to test his loyalty.

From the 4671 story library. Read the story below, or explore its original reader and connected world.

Howard and the Habidash

At thirty-five, Howard joined the Habidash because Mara, their child Eli and his mother June depended on money his irregular waste-pump repair shifts could no longer provide.

They lived in Habitation Tier 18, called Wet Eighteen by the people who had to live there. Condensation and water-reclamation pipes leaked through ceilings and service walls faster than the maintenance crews could patch them. Mara, thirty-three, tested reclaimed water under a low-status laboratory classification. Eli was eight. June was sixty-one and had spent most of her working life inspecting and patching condensation lines. When reduced shift eligibility cut June’s income, the household leaned harder on Howard. No illness had done this to them. New Atriad’s classifications had simply decided how much work each of them was still allowed.

Their two-room unit was cramped but cared for. It sat against a service wall with access panels that never fitted properly, catch basins beneath the worst leaks and fold-down beds that returned a little floor space each morning. Their work clothes dried beside the warmest pipe because it was the only reliable heat in the room.

Housing and work classifications passed from one generation to the next, binding families to water processing, pipe repair, waste systems and other low-status work across the neighbouring tiers. Severe poverty remained part of ordinary life despite the wealth and power of the lunar capital.

Howard had watched every honest option narrow until rumours of the pirates began to sound less like warnings and more like work. Habidash crews crossed the Sol system taking cargo, ships and lives. They were dangerous, but danger appeared negotiable when hunger was not.

He told himself he would endure them for his family.

That lie carried him aboard.

At first, Howard tried to look only at the part that concerned him. A task was given. He completed it. A superior spoke. He obeyed. The ship moved through the dark between settlements, and Howard counted the distance as though it separated him from responsibility.

It did not.

The Habidash did not simply steal what they needed. They took what they wanted because no one nearby could stop them. Cargo crews were threatened after surrendering. Settlements were damaged after payment had already been made. Violence continued long after it had served any practical purpose.

Howard saw enough to understand that cruelty was not a failure of discipline. It was part of how the pirates kept discipline.

By then, leaving was no longer a decision he could make alone. He knew their faces, their habits and the routes they used. They knew where he had come from and why he needed them. Every day aboard made him more useful to the Habidash and less able to return to the person his family believed him to be.

Then they captured the cargo pilot.

The pilot was innocent. Howard knew that before anyone ordered him forward. There had been no battle worth avenging and no hidden threat to uncover. The prisoner was simply available, and the Habidash had decided Howard needed to be blooded. The killing would test his obedience and make his guilt useful to them.

He was made to take part in killing the pilot while the crew watched.

Howard would remember the pilot afterward in fragments. He would remember the enclosed air, the people watching and the certainty that refusing would not save the prisoner. He would remember that he obeyed.

Whatever the Habidash called the act, it left Howard unable to pretend he was merely working among violent people. Their crime now included him.

He carried it through every passage of the ship. He tried to imagine telling his family and could never reach the end of the confession. He could not explain that he had joined for them without making them part of his excuse. He could not ask for forgiveness while the pilot remained dead. He could not return home and place what he had become at their door.

The pirates had trapped him more completely than any locked compartment could have done. Even escape would not restore the life he had entered with.

Howard began watching his superior.

The plan that formed was not complicated. There was no route out of the Habidash that he could see, so he chose the person directly above him and decided to kill them. Perhaps it was revenge. Perhaps it was punishment. Perhaps Howard needed one final act that belonged to him rather than to the people who had blooded him.

He knew it would not undo the cargo pilot’s death. He did not believe it would make him innocent. It was only the last direction left that did not feel like obedience.

When Howard acted, the clean moment he had imagined disappeared at once.

Gunfire filled the ship. The narrow space gave every shot a hard mechanical violence. Howard fired, moved and lost track of who had seen him first. The Habidash answered from more than one direction. His attempt became noise, impact and the sudden knowledge that he had never controlled what would happen next.

A round struck him in the head.

His body fell before his thoughts did.

In the last moments available to him, Howard did not think about the pirate he had tried to kill. He thought about Mara, Eli and June in Wet Eighteen and everything he had failed to tell them. He imagined the Habidash reaching them. He imagined slavery, exploitation and the worst markets hidden across the system. Fear turned every possibility into certainty, and he died believing that his attempt to provide for them might instead have exposed them to ruin.

He never learned whether that fear was true.

He never returned to confess what he had done, and he never found a way to repair it. The last thing the Habidash took from Howard was the chance to know what became of the people for whom he had joined them.

Photography by Tomas Heligr-Pyke · Bokeh

Two private diaries follow the same long-distance relationship from the person being recruited and the Follower guiding him.

From the 4671 story library. Read the story below, or explore its original reader and connected world.

Evan’s Diary

First Entry

I have a girlfriend.

That still looks impossible when I write it down, but it is true. Her name is Sophia. She lives in New Hope City and I live in Nybell, so for now we are voices, messages and faces behind glass. I thought the distance would make it feel less real. It does not.

We have made a plan. In three years, we will finally be together.

Three years sounds enormous when I say it aloud. Sophia says it is nothing compared with forever.

I believe her.

The Distance

We are alike in ways I have never been alike with anyone.

Sophia knows what I mean before I finish saying it. When I cannot sleep, she is awake. When I feel stupid, she makes the feeling sound temporary. Sometimes we sit without speaking and listen to each other breathe through the call. I used to think silence meant there was nothing left to say. With her, silence feels like proof that neither of us has gone.

I love her. There is no cleverer way to write it.

I want to live with her. I want to wake beside her and know there is no screen between us. She wants that too.

She keeps saying forever.

I have started saying it back.

What She Believes

Sophia told me about her beliefs tonight.

She would not give me a name for them. She said names are what outsiders use when they want to put a wall around something. Her people believe there is more beyond us: something greater than one life, one body or one world.

At first it sounded like every religion I had ever heard of. I told her that. She laughed, but she did not get angry.

She said ordinary religions ask people to believe without showing them anything. Her people have proof.

I asked what the proof was.

She said I was not ready.

I thought that would annoy me. Instead, I wanted to be ready.

The Blackout

I could not reach Sophia.

The calls failed first. Then the lights went out, followed by every screen in the building. No one could agree how far the blackout reached. Each station that returned named somewhere different.

When the news feeds came back, they showed a city hanging above another country.

Some reports said Nussia. Others said Uchen or Atriad. The footage never looked the same twice. Towers stood upside down against the clouds. Roads ended in open sky. For a few seconds, the shape above them moved like a reflection after the thing being reflected had gone.

People online called it a hoax. Officials called it an atmospheric fault. Sophia called me when the network returned and asked what I thought I had seen.

I told her I did not know.

“Proof,” she said.

She would not explain further.

Purpose

Sophia has been teaching me slowly.

I understand why now. If she had told me everything at once, I would have heard the words without understanding them. A body believes it belongs to itself. A mind believes it is alone. Work, money, hunger, sleep: all of it keeps a person staring down at the floor when there is something above them, around them and waiting beyond them.

I used to wake, work and come home so tired that I could barely speak to the one person I loved. I called that a life because everyone else did.

I have found my purpose.

A life without purpose is not a life.

I do not need to go to work anymore. The Great One will provide what matters.

Sophia says I am close.

Offerings

I made my first offerings.

The animals were small. I bought them from people who asked no questions. Their lives would have ended eventually. Everything made of flesh ends. The only choice is whether the ending means something.

I thought I would feel shame. I felt fear, then stillness.

Sophia stayed with me through the call. She told me not to look away. She said an offering made in shame is an offering refused.

When it was done, she told me she was proud of me.

No one has said that in a long time.

The Followers help us see what the rest of the world refuses to see. Flesh is not sacred because it lasts. Flesh is sacred because it can be given.

The Cube will guide us to sanctuary.

One

We are all one.

And one is all.

Our lives are joined.

Our stories are shared.

What we surrender will remain.

What remains will change what follows.

We will worship.

We will follow.

The Great Ones are beyond us.

The Artefact is before us.

The Cube will open the way.

Last Entry

Sophia is here.

I collected her from the airport this morning. For one terrible second I thought we would not recognise each other without a screen between us. Then she smiled, and there was no distance left.

We came back to my apartment and prepared the room. The table stands in the centre. The chairs face away from it. We pressed our fingers until blood rose, then drew the Cube together across the tabletop. Her lines met mine at every corner.

Soon, each of us will open the other. Flesh and blood will join what distance kept apart. There will be pain, but pain is only the body begging to remain separate.

Sophia says we will be together forever.

I believed her when forever meant three years.

I believe her now.

This is the last thing I will write before we begin.

Sophia’s Diary

First Entry

I have a boyfriend.

His name is Evan. He lives in Nybell, far enough away that we cannot meet yet. That helps. The cause cannot be hurried. If I tell him what waits for us before he understands me, fear will answer before love can.

He is not a believer.

That will change.

When the time comes, we can complete the cause together. Neither of us will have to cross alone.

Patience

We have agreed to wait three years before we live together.

Evan thinks the wait is practical. He talks about money, work and all the ordinary arrangements people make when they believe a future is something they own.

I let him talk.

I listen to everything. That is how I learn which parts of him are empty and which words will fit inside them.

The emotional connection between us is real. I can feel the urge to love him.

He is amazing.

The Teaching

It is time to begin converting Evan.

I started with the idea of something higher. Nothing about the Followers. Nothing about the blood. Nothing about what a body must become before it can be left behind.

He called it religion. I laughed because he expected me to. Then I told him we have proof.

That was enough.

Curiosity opens more gently than fear. Once opened, it is difficult to close.

I teach him slowly. I do not want to lose my one and only other half.

First, I teach him that the body is temporary.

Then, that temporary things can still be used.

Then, that sacrifice is not destruction when something greater receives it.

Soon I will tell him about the Cube.

The Failed Opening

Another cell tried to force an artefact awake.

They believed they could feed it enough power to open a passage. Instead, the grid failed and two places occupied the same sky. The news called the other place a floating city because ordinary language has no words for one world showing through another.

The reports cannot agree whether it appeared over Nussia, Uchen or Atriad. Perhaps the overlap was larger than anyone understood. Perhaps frightened people gave the same impossible sight three different names.

The cell has not spoken since.

Evan saw enough to become curious. I told him it was proof and let the silence do the rest.

Failure can still open someone.

Conversion

He has stopped going to work.

He says the Great One will provide. He says it with the certainty of someone repeating his own thought, not mine. That is how I know the teaching has settled inside him.

He made his first offerings. I watched. He did not turn away.

Evan is now one of us.

I thought triumph would feel louder.

Instead, I listened to him breathing after the call and wished I could touch his face.

One

We are all one.

And one is all.

Our lives are joined.

Our stories are shared.

What we surrender will remain.

What remains will change what follows.

We will worship.

We will follow.

The Great Ones are beyond us.

The Artefact is before us.

The Cube will open the way.

Last Entry

Evan met me at the airport.

He looked exactly like himself and nothing like the person on the screen. When he held me, I felt the distance collapse all at once.

The room is ready. He remembered the table and the chairs. I did not have to correct him. We marked the tabletop together, his blood joining mine until the Cube was whole.

Soon we will open each other. Our flesh will end. What is joined beneath it will continue.

Evan believes we are doing this because we love each other.

I chose him for the cause before I knew him.

Now, when I try to separate what I feel from what we are about to do, I cannot find the line.

There is no distance left.

Photography by Tomas Heligr-Pyke · Bokeh

An unnamed detective follows an anonymous file into the cover-up surrounding Annabelle Klanders’s death.

From the 4671 story library. Read the story below, or explore its original reader and connected world.

The Room

I woke with my wrists chained to a chair.

The room was too dark to measure. A line of light beneath the door gave me the floor, my shoes and the first few links hanging from each arm. Everything beyond that line might as well have been open space.

I pulled once. The chair moved less than an inch.

Two days earlier, I had been in my office wondering how long I could keep pretending it was a business. Then an unfamiliar file appeared on my desk.

That was all it took.

One file. One dead student. One question I should have left unanswered.

Two Days Earlier

The folder had no return address, invoice or client name. Inside was a short note.

Annabelle Klanders died one week ago. They called it suicide. Something is wrong with the body.

The rest was copied from official records.

Annabelle was a young university student. Her computer had been wiped, and the storage inside it had been physically destroyed. She had been found at home with wounds across her wrists and a bloodied razor left beside her.

The authorities had accepted the scene.

Someone else had placed the file on my desk.

I should have taken it to the police. I should have thrown it away. Instead, I looked at the unpaid work stacked beside it and decided that an anonymous client could still become a paying one.

I began with the people who knew her.

They disagreed about almost everything except the conclusion.

Annabelle had been kind. Annabelle had been difficult. Annabelle kept to herself. Annabelle spoke to everyone. She was serious about her studies. She had been distracted. She seemed happy. She did not seem happy.

But no one expected her to die.

Clive was the last interview.

He sat opposite me with his hands locked together and stared at the space over my shoulder.

“She was my friend,” he said. “We were messaging on the day it happened.”

“About what?”

“People she wanted to meet.”

“What people?”

His hands tightened.

“Some call them rebels. The ones with the cuts.”

“Cuts where?”

“Their hands. Their fingernails. I do not know.”

“Was Annabelle joining them?”

Clive looked at me then. Whatever answer had been ready disappeared before he spoke.

“I do not know. She stopped replying.”

“You were talking to her on the day she died. What did she say before she stopped?”

He stood.

“That is all I know.”

It was not all he knew. I could hear that much. I let him leave anyway.

The Followers

The description led me away from political groups and into old rumours: private rites, artefacts, scars hidden beneath ordinary clothes, deep arrow-shaped cuts carved into fingernails.

The Followers.

Most references contradicted one another. Some called them a church. Others called them a criminal network, an invented enemy or an excuse for unexplained bodies. The only useful name I found was Zander.

I sent a request through a contact channel and promised anonymity. His reply named a park and a time.

He was already on the bench when I arrived.

“Sit,” he said. “Look ahead. Do not look at me.”

I sat at the other end of the bench. The path in front of us was empty.

“You know why I am here?” I asked.

“Annabelle Klanders.”

“Clive said she was trying to meet rebels.”

“They are not rebels.”

“Then what are they?”

“Followers. They do not care who governs your city. They care about things older than it.”

“Was Annabelle one of them?”

“She was trying to reach them. That is not the same thing.”

“Why?”

“You are asking the question that got her killed.”

I turned toward him without thinking.

“Look forward,” he said.

I faced the path again.

“Clive left something out,” Zander continued. “Annabelle had been asking questions about Iron Fist. Not stories. Names, decisions, things that were not meant to remain written down. Enough of what she found was true.”

“What does that have to do with the Followers?”

“Maybe she thought they had evidence. Maybe she wanted protection. Maybe she wanted to become one of them. It no longer matters. Iron Fist knew she was looking.”

“You are saying they killed her.”

“They followed her home. They destroyed what was on her computer and made the room tell a simpler story.”

The wounds. The razor. The clean official conclusion.

“Can you prove that?”

“No. Neither can you.”

“Then why tell me?”

“So you understand what happens next.”

Zander stood.

“Close the file. Forget Annabelle. Forget my name.”

“You said the Followers did not kill her.”

“I said Iron Fist did. I did not say the Followers were safe.”

He started down the path.

“Is that all?” I called.

He did not turn.

“It was already too late for Annabelle. Decide whether it is too late for you.”

The Car

I walked in the opposite direction and told myself the case was finished.

There was no evidence I could use. No client to report to. No witness willing to sign a statement. I had a destroyed computer, a frightened friend and the word of a man who would not let me look at him.

I reached the end of the park.

A car door opened behind me.

Someone caught my arms. Another hand covered my mouth. The pavement vanished beneath my feet, and the inside of the car swallowed me before I managed to shout.

Then came the room, the chair and the chains.

The Offer

The door opened.

Light crossed the floor and climbed the legs of my chair. A man entered, shut the door behind him and stopped where I could not see his face.

“You spoke to a Follower,” he said.

“I spoke to a source. I am a detective.”

“You spoke to a Follower about Annabelle Klanders.”

“I was hired to investigate her death.”

“By whom?”

“I do not know. The file was anonymous.”

He stepped closer. A blade touched the skin beneath my jaw.

“That answer makes you useless or dishonest. Neither keeps you alive.”

“I am telling you the truth.”

“Are you one of them?”

“No.”

“Did Zander give you names?”

“No.”

The blade pressed harder.

“What did he tell you about us?”

I thought of Annabelle’s destroyed computer. The wounds placed where people expected to find them. The razor completing the picture.

“Nothing I could prove,” I said.

For the first time, the man laughed.

“That is the correct answer.”

He lowered the knife but did not move away.

“You cannot return to your office knowing what you know. We can leave you here, or you can make yourself useful.”

“You want me to join Iron Fist.”

“I do not care what you believe. You will investigate when instructed. You will report what you find. You will speak to no one about this room, Zander or Annabelle Klanders. The details come after you leave.”

It was not an offer. It was a method of deciding whether I died in the chair or somewhere later.

“And if I refuse?”

The knife returned to my throat.

I closed my eyes.

“I accept.”

The blade lifted.

I had not joined them. I had chosen the only answer that kept me breathing.

Whether there was still a difference, I would learn when they gave me the next file.

Photography by Tomas Heligr-Pyke · Bokeh

A camera fixed in the back seat records Jain and Alex’s attempt to leave an Iron Fist puppet state before curfew.

From the 4671 story library. Read the story below, or explore its original reader and connected world.

Screen Rule

The camera never leaves the back seat of the car.

It may see through the windows, between the front seats, or into the mirrors, but it never follows anyone outside. The audience learns what happens beyond the doors through partial movement, reflected light, sound, and what does not return.

Border

BLACK.

Rain taps against metal.

A car unlocks. The interior light comes on, revealing an empty back seat and two vacant places in front. The passenger door opens first. JAIN gets in and immediately turns towards the driver’s side.

ALEX enters carrying a document wallet under his coat. He closes the door carefully instead of slamming it.

For several seconds, neither of them starts the car.

JAIN: You have everything?

Alex opens the wallet. Two passports. Two travel permits. A folded vehicle authorisation.

ALEX: Everything he sold me.

JAIN: That is not what I asked.

ALEX: It is everything they ask for at the border.

Jain looks through the windscreen. Rainwater bends the streetlights into long white streaks.

JAIN: And if they ask for something else?

ALEX: Then I give them something else.

JAIN: Alex.

He stops arranging the papers.

ALEX: We are leaving tonight. We cross before curfew. By morning we are outside their records.

JAIN: Their records go everywhere.

ALEX: Not everywhere.

JAIN: That is what people say before Iron Fist arrives.

Alex reaches across the console and takes her hand. Jain guides it down to her stomach.

ALEX: I will keep you both safe.

JAIN: Do not promise what you cannot control.

ALEX: I promise I will not stop trying.

Jain nods. It is not agreement, but it is enough for him to start the engine.

The interior light fades.

They drive.

For a while there is only rain, the wipers, and the wet road passing beneath the headlights.

JAIN: Where do we go after the crossing?

ALEX: Inland first. Then south, if the road is open.

JAIN: You have never been there.

ALEX: Neither have the people who told us it was safe.

JAIN: That does not help.

ALEX: I know.

He squeezes her hand once and lets go to change gear.

ALEX: We need fuel. Food too. The first town across the line might not take our money.

JAIN: The first town across the line might send us back.

ALEX: Then we keep driving.

He turns on the radio.

A calm official voice fills the car.

RADIO ANNOUNCER: Border restrictions remain in force. Travel after twenty-three hundred hours requires valid identity documents, movement permits, and registered destination authority.

Jain looks at the dashboard clock.

RADIO ANNOUNCER: Citizens who have knowingly communicated with a Follower, assisted a Follower, or withheld information concerning Follower activity are ordered to report to the nearest community centre for dismissal.

Alex turns the volume down but does not turn it off.

ALEX: They put Jimmy on the list.

JAIN: They accused Jimmy. That is not the same thing.

ALEX: It is to them.

RADIO ANNOUNCER: Curfew begins at twenty-three hundred hours. Violators will be detained. Protect your home. Protect your nation. Obedience preserves peace.

Alex switches the radio off.

ALEX: If they took him, they will read every name attached to his.

JAIN: We were already leaving.

ALEX: I should have gone to him.

JAIN: And then there would be three names on the list.

ALEX: There may already be.

Jain watches him. He keeps his eyes on the road.

JAIN: What matters now is getting through.

Alex nods, but his jaw remains tight.

The car turns beneath the bright canopy of a roadside fuel station. Only one other vehicle is present, parked without lights near the building.

Alex takes the keys.

ALEX: Lock the doors while I am inside.

JAIN: You said you would not leave me.

ALEX: Thirty seconds.

JAIN: You do not know how long thirty seconds is tonight.

He leans over and kisses her.

ALEX: Then count. I will be back before you finish.

Alex gets out. Jain locks the doors.

She watches him cross in front of the car and disappear into the station.

One second.

Two.

Three.

She stops counting at twelve when someone knocks on the passenger window.

A STRANGER stands outside in the rain. Their face is distorted by water on the glass.

Jain lowers the window by a few centimetres.

JAIN: What do you want?

STRANGER: Curfew is close.

JAIN: We know.

STRANGER: Then go home.

JAIN: We are going home.

The stranger looks at the document wallet on the driver’s seat. Jain covers it with her hand.

STRANGER: Not by the western road, you are not.

Jain says nothing.

STRANGER: They changed the checkpoint shift this afternoon. New officers. They are checking the paper against the register now.

JAIN: Why are you telling me?

STRANGER: Because somebody once told me.

Alex returns carrying a small paper bag and two bottles of water. He stops when he sees the stranger.

ALEX: Is there a problem?

STRANGER: There will be if you keep driving.

The stranger walks away. They do not enter the waiting vehicle. They disappear past the edge of the rear window.

Alex gets in and locks the door.

JAIN: He knows.

ALEX: He guessed.

JAIN: He said they are checking the register.

ALEX: The papers are registered.

JAIN: To people who are not us.

Alex puts the food on the back seat beside the camera. Water, cereal, sealed packets, whatever could be eaten without stopping.

ALEX: We cannot go back.

JAIN: I know.

ALEX: We can wait until tomorrow.

JAIN: Curfew officers will check the car before morning.

ALEX: I know.

Neither of them says what remains.

Alex starts the car.

The station lights fall behind them.

JAIN: Alex.

ALEX: I am here.

JAIN: Whatever happens, do not make me face it alone.

ALEX: I am here.

This time he does not promise anything else.

The road narrows. Wire fencing appears beyond the passenger window. Floodlights grow brighter ahead.

Traffic has been divided into empty lanes by concrete barriers. A black banner hangs from the checkpoint building, wet cloth pulling around one plain red fist.

Alex slows beside a booth.

A gloved hand knocks twice on his window.

He lowers it.

BORDER GUARD: Papers.

Alex gives over the document wallet.

The guard turns away. Through the rain-streaked rear window, a second officer walks behind the car and examines the plate.

Jain’s hand finds Alex’s beneath the level of the window.

Five seconds.

Ten.

The guard returns without the wallet.

BORDER GUARD: Engine off. Driver, step out.

ALEX: Is something wrong with the permits?

BORDER GUARD: Step out.

Alex looks at Jain.

ALEX: Stay in the car.

He opens the door.

Hands seize his coat before he can stand. The driver’s door is pulled wide, filling the back-seat view with rain and white checkpoint light.

Alex disappears beyond its frame.

ALEX: Wait. She is pregnant. Please, just let her go back.

A strike lands outside.

Jain reaches across the console, but the empty driver’s seat blocks most of the view.

Another strike.

Alex falls low enough for one hand to appear on the wet road beneath the open door.

BORDER GUARD: False identity. False movement permit. Association flag.

ALEX: We have done nothing.

BORDER GUARD: You tried to leave.

A flash turns every window white.

The gunshot reaches the back seat an instant later.

Alex’s hand stops moving.

Jain screams his name and claws at her door lock. It opens from outside before she can reach it.

The passenger door swings wide.

JAIN: Why did you do that? He was not fighting you. Why did you do that?

A guard’s arm reaches into the car. Jain grips the seat belt with both hands.

JAIN: Please. I am pregnant. Please.

The belt releases.

Jain is dragged from the passenger seat. Her fingers hold the door frame for a moment, then slip away.

The camera sees only the open door, hard rain, and the lower half of the checkpoint barrier.

JAIN: Alex?

A second flash.

A second gunshot.

Silence, except for the rain and the idling engine.

An officer leans into the car, turns the ignition off, and removes the false documents from the console. Their face never enters the frame.

The doors are left open.

Rain begins to darken both empty front seats.

CUT TO BLACK.

Photography by Tomas Heligr-Pyke · Bokeh

An unnamed survivor finds passage off a dying Earth, but the Ark carries a crime that New Atriad will turn into history.

From the 4671 story library. Read the story below, or explore its original reader and connected world.

The Third Day

I used to think being the last person alive would feel like silence.

It did not.

It sounded like alarms with nobody left to answer them. It sounded like water running in apartments above empty shops. It sounded like a dog barking behind a locked door until one morning it stopped. At night, when the power failed district by district, the city settled with long cracks and groans, as though the buildings were trying to speak to each other after we had gone.

The panic began before the disease had a name.

People crowded the stations. They emptied pharmacies and fought over bottled water. The government told us to stay indoors, then told us to move toward the shelters, then denied it had ever ordered an evacuation. Iron Fist broadcasts replaced every channel. A calm voice promised that the situation was contained while armed vehicles sealed the roads out of the city.

Then the bombs fell beyond the river.

I spent the first night on the floor of a stranger’s laundry with six other people. We listened to impacts roll through the ground and watched dust jump from the ceiling. One man prayed. Another kept checking a dead phone. Near dawn, someone began coughing in the hall.

By morning, the laundry was empty except for me.

I never learned whether the others fled or were taken. There was blood on the tiles and a trail of it through the rear door, but no bodies. That became ordinary. I opened houses and found meals left on tables, taps running, beds stripped apart. I found blood in handprints and long smears where people had been dragged. I found no one who could tell me what had happened.

On the third day, I saw the Arks.

They rose from the launch fields north of the city, one after another, slow at first and then impossibly fast. Their engines turned the clouds orange. Even from kilometres away I could see the black Iron Fist banners hanging from the towers, each marked with a single red fist. The cloth snapped in the pressure waves while the ships climbed above us.

They had room to leave.

They left us anyway.

I stood in the street and watched until the last engine became a white point. That was when people began running from the bomb district.

I called them mutations then. Everyone did. We believed the craters had changed them. We believed one touch meant death. Fear gave us rules because rules made the end of the world easier to understand.

What came down the street was worse than anything fear had prepared me for.

They were people. That was the part none of us wanted to see.

Some had gone grey around the mouth and eyes, their skin stretched tight over swollen veins. Others were already coming apart. A woman’s cheek had split from the corner of her mouth to her ear, exposing teeth lacquered dark with blood, yet her eyes moved from doorway to doorway with terrified attention. A man ran on a broken ankle. The joint rolled sideways each time his foot struck the road, bone pressing white against skin that refused to tear. He screamed with every step and kept running.

Their wounds did not behave like wounds. Skin loosened in wet sheets. Fingers wore down to shining knuckles where they had clawed at concrete. Jaws worked after lips had split. They snapped at each other, at doors, at their own arms. Between those fits they cried, begged, called names, or stared as if they were trapped behind their own faces.

One of them saw me.

For a moment she stopped.

“Please,” she said.

Then her body folded forward and charged.

I hid in a storage cupboard inside the nearest building. She struck the door until the frame shook. Others joined her. I could hear nails scraping down the metal, then the soft knock of bare finger bones. I held a shelf against the handle and waited for them to find another sound.

Hours later, they moved on.

The next morning I left the city.

The Long Walk

I went where there were fewer people.

Forests first, then farms, then the dry country beyond them. I avoided highways and slept in culverts, sheds and houses set far back from the road. I drank from tanks when the water did not smell wrong. I learned which tins could swell without leaking and which roots would make my tongue numb. When I found batteries, I saved them for the radio. The radio gave me static, old evacuation orders and Iron Fist voices telling survivors to report to collection points that had already burned.

I met people sometimes.

Once, I travelled for four days with a mother and her adult son. The son had a cut under his sleeve that he would not show us. On the fifth morning he woke shaking so hard his teeth broke against each other. His mother held him while blood ran from his nose and ears. He kept saying he was still himself.

He was.

That did not stop his hands from closing around her throat.

I ran before I saw how it ended.

After that, I kept my distance from everyone.

The months became weather. Rain inside my boots. Heat shimmering above empty roads. Cold that made abandoned houses sound occupied as timber contracted in the walls. I stopped counting days when I lost my pack crossing a flooded creek. I know at least a year passed because I saw the same kind of flowers bloom twice.

Loneliness changed the shape of things.

In a department store, I sat six mannequins around a table and gave them names. In another town, I found a body seated behind the counter of a service station and told him where I had been. He listened for three nights before the smell forced me to leave. I apologised when I went.

I spoke aloud because silence made every thought sound final.

Then I found the settlement.

It stood inside an old Iron Fist service field, hidden behind collapsed hangars and concrete walls stained black by engine exhaust. Smoke rose from cook fires. Washing hung between antenna masts. There were guards on the gate and children chalking pictures on the tarmac.

Living children.

One of them stared at me through the bars. I reached between them before the guard could stop me and touched the child’s face.

Warm skin. A pulse under the jaw. Breath.

I began crying so hard I could not tell them my name.

They let me in after three days of isolation.

There were more than a hundred people inside the walls. Farmers, labourers, nurses, teachers, mechanics and families who had followed fragments of an evacuation signal. At the centre of the field lay an Ark on its side supports, too large to understand as a vehicle. Dust buried its lower hull. Scaffolds climbed its flanks. A black panel near the bow still carried the red fist.

Iron Fist had shut the vessel down before the last launch wave. Some said the assigned crew had been recalled to a better ship. Others said they had died in quarantine. Nobody knew. The settlement had found the Ark empty, damaged and still registered in the evacuation network.

They were repairing it.

For the first time since the city, I believed I might live.

The Ark

Everyone worked.

We stripped wire from buried offices and hauled fuel cells from a military depot. We cleaned water tanks, packed seed stores and sealed the habitation pods stacked beneath the Ark’s belly. I crawled through service ducts because I was thin and no longer afraid of dark spaces. I knew how to ration food, find leaks, sleep through alarms and keep moving when something smelled dead. Those were useful skills now.

The people who directed the repairs had been engineers, medical officers, administrators and licensed Ark contractors before the collapse. They kept the only working terminal in a locked room. They wore old credentials on cords around their necks even though there was no government office left to inspect them.

They said the credentials would get us through the orbital grid.

Their leader had an Ark Key.

I never saw it clearly. I saw only a dark object placed against the command console and the Iron Fist system waking around it. Lights passed across the bridge. A recorded voice welcomed an authorised crew. The room cheered.

After that, we called the Moon our destination.

Some people called it a new world. Others called it a utopia. I had watched Iron Fist ships abandon a city, so I did not believe in either word. I believed in sealed air and distance from the infection. That was enough.

The work took months.

As launch approached, the technical group began weighing everything. Water. Seed. Tools. Fuel. Oxygen. People.

The leader stopped saying everyone would fit.

He spoke instead about survival margins. He divided us into crews and dependants. He asked what each person could contribute once we reached the Moon. The lists changed every night.

No one wanted to ask what happened to a name when it disappeared.

On the final evening, we ate together beneath the Ark. Someone opened the last bottles from a ruined hotel. Children ran between the tables while adults argued about what they would build first. A woman beside me said she wanted a room with a window facing Earth.

I told her I never wanted to see Earth again.

She laughed because she thought I was joking.

We slept in the old barracks, ready to board before dawn.

I woke to a muffled shot.

Then another.

For several seconds nobody in my room moved. A man in the bunk above me whispered that it might be infected outside the wall. The door opened before he finished speaking.

Two members of the technical crew entered wearing dark maintenance coveralls. One carried a pistol. The other held a list.

The first shot took the man above me through the chest. He fell halfway from the bunk and hung there, one arm striking the frame. My roommate sat up. The second shot threw her against the wall.

I rolled beneath the lowest bed and pressed myself into the dust.

Boots crossed the floor.

The person with the list read names. The gun answered each one. Some people never woke. Some woke enough to plead. A child began screaming in the next room, one thin sound cut short by a shot and then another.

Blood crept under the bed toward my face.

I held my breath as it touched my sleeve.

The boots stopped beside me.

For one second I was certain they could hear my heart striking the floor.

Then someone called from the corridor. The boots moved away. The door closed.

I stayed beneath the bed until the shooting spread to the far barracks.

Then I ran.

The Hatch

The launch field had become a place of shadows and muzzle flashes.

People stumbled between the buildings in night clothes. Some tried to reach the Ark. Others ran for the walls. The technical crew controlled the boarding ramp and shot anyone whose name was not on their list.

I climbed through a maintenance opening beneath the aft hull. I had sealed it myself two days earlier and knew the bolts were only hand-tight. Inside, I crawled past fuel lines and dropped into the cargo hold between containers of dry food.

The inner hatch locked behind me.

The engines began their start cycle.

Outside, the surviving settlers realised what was happening. They rushed the main hatch. I heard fists and tools striking the outer door, a frantic metal thunder that travelled through the hull.

“Open it!”

“My daughter is inside!”

“Please!”

The loudspeaker ordered them to clear the launch area.

They kept pounding.

Beyond them came another sound: many feet moving across the tarmac.

The engine tests and gunfire had drawn the infected from the surrounding towns. They poured through a breach in the perimeter wall, dozens at first and then hundreds, bodies colliding and climbing over one another. In the red launch lights I saw faces stripped to tendon, arms hanging from opened shoulders, torsos wrapped in old bandages turned black and rigid. Some ran with the desperate speed of the newly infected. Older ones dragged ruined legs or crawled with fingers worn down to bone.

They were moaning, but words moved through it.

Hungry.

Help me.

Run.

The people at the hatch looked back. Some pressed themselves against the Ark. Others tried to fight through the horde. A man lifted a little boy above his head as if someone inside might reach down and take him.

I was inside.

I could have opened the service release from the cargo hold.

I knew where it was.

I did nothing.

The Ark launched while they were still striking the hull.

Acceleration crushed me between the containers. The hammering disappeared beneath the engines. Through a narrow inspection port, I watched the field fall away: the fires, the breached walls, the living and infected becoming one moving stain around the empty launch cradle.

Then cloud covered everything.

New Atriad

I hid for nearly two days.

There was enough food in the cargo hold to feed a city branch, and I still rationed every mouthful. Habit had become stronger than hunger. I slept between containers and woke whenever the ship changed course.

On the second night, armed crew found me.

They dragged me to the bridge and forced me onto my knees. The leader stood above me with his old credential still hanging from his neck. On the forward display, the Moon filled half the sky.

He asked how I had boarded.

I told him.

He asked what I had seen in the barracks.

I said I had seen nothing.

He knew I was lying. I knew he knew.

“What can you do?” he asked.

I told him about the service ducts, the water systems, the wilderness, the months alone. I told him I could live on almost nothing and keep machinery working with less.

He looked at me for a long time.

“Then you survived,” he said.

It was not mercy. It was employment.

Before we reached the lunar perimeter, he made me repeat the account he had prepared. Infection entered the settlement. Panic followed. The outer camp was lost in a riot. The authorised crew recovered the vessel and launched with every viable survivor.

I repeated it until I could say it without pausing.

Iron Fist defence platforms found us before we saw New Atriad. Targeting warnings filled the bridge. The crew transmitted the Ark’s identity and the stolen Key’s authorisation. For several minutes, nobody breathed.

Then the weapons released their lock.

Escort craft took position around us.

Iron Fist did not welcome us as founders. It processed us.

Officers separated us by credential, health, occupation and usefulness. They detained me because my name was not on the manifest. The leader gave them the story about infection and riot. I confirmed it. The officers looked at the intact Ark, the packed equipment and the technical crew who could make it work. No one asked a second question.

The habitat pods were lowered into a branch zone beneath New Atriad. Iron Fist already owned the ground, the air, the power and the locks. The technical crew received quarters and administrative work. I received a temporary number and a place in maintenance.

The city grew around us.

We sealed tunnels, opened mines and built chambers for water and crops. New families filled the pods. Artificial gravity pulled at our bones. Children who had never stood beneath an open sky learned that Earth was a bright object behind reinforced glass.

Years later, a plaque appeared in the transit hall. It said our Ark had completed an emergency rescue after its launch site was overrun. It praised the authorised crew for saving essential personnel and equipment.

People stopped to read it.

The leader attended the unveiling.

I repaired an air recycler in the wall behind them.

The Door

New Atriad calls us survivors.

It is a useful word. It tells what happened to us without asking what we did.

I survived the city. I survived the wilderness. I survived the barracks because I hid beneath a bed while other people were killed. I survived the launch because I stayed inside while families struck the hatch. I survived the Moon because I repeated a murderer’s lie.

The woman who wanted a window facing Earth did not survive.

The father holding up his son did not survive.

The people whose blood ran beneath my bed did not survive.

I did.

For years I told myself that surviving meant I could repay them. I kept air moving through the branch. I repaired pumps before crops died. I taught apprentices how to hear a failing seal. People live because of work I have done.

It has never silenced the hatch.

I still hear the pounding when the ventilation changes speed. I hear voices in the pipes. Sometimes I wake certain that there are people outside my door, asking me to open it before the infected reach them.

There is a plaque in the transit hall now. There are school lessons about the rescue. The leader’s lie has become history, and my silence is built into it.

I cannot pull the Ark back to Earth.

I cannot open that hatch.

There is only one door left for me.

The inner lock closes behind me. Through the small window, New Atriad stretches across the dark: lamps, towers and pressurised tunnels made from everything we carried away. Earth hangs above it, blue enough to look untouched.

The outer seal begins to turn.

This time, I am the one outside.

Photo: Bokeh / Tomedia. Silhouettes on the Beach at Night.

My notes on prayer contain a very simple suggestion: please, thank you and sorry.

I like that. I am perfectly capable of making a simple thing unnecessarily complicated, and prayer gives me plenty of opportunities. Am I saying enough? Am I saying the right thing? Have I understood what God wants before I have even started speaking?

Meanwhile, there are things I need help with, things I am grateful for and things I need to own up to. Those seem like reasonable places to start.

A conversation without an audience

In Matthew 6:5–8, Jesus talks about people praying to be seen. He tells his listeners to go into their room, shut the door and pray to their Father in secret. He also warns against piling up words as though the quantity will make God listen.

But thou, when thou prayest, enter into thy closet, and when thou hast shut thy door, pray to thy Father which is in secret; and thy Father which seeth in secret shall reward thee openly.

Matthew 6:6 (KJV)

That takes quite a lot of the performance out of it.

Nobody needs to hear how eloquently I can explain my problems. God already knows what I need. I can speak plainly and give him my attention.

One of the sermons spent time on the room itself. My notes veer off into the Greek word and descriptions of storehouses. But the instruction I need to act on is clear enough in English. Go somewhere private. Close the door. Pray.

I don’t need to finish a word study before I can manage that.

Something will fill the time

In another set of notes I wrote about ordering our private lives. Unless we make room deliberately, other things will take it.

I know how easily that happens. Work, plans, things I want to build, whatever thought I have decided needs pursuing immediately. By the time I have dealt with one thing, I have usually found several more.

Jesus’ pattern in Mark 1:35 is to withdraw and pray even when people want his attention. There are genuine needs around him. The time with the Father still happens.

And in the morning, rising up a great while before day, he went out, and departed into a solitary place, and there prayed.

Mark 1:35 (KJV)

For me, the practical step is giving prayer an actual place in the day. Somewhere I can stop opening another thing. A few minutes that I have already decided belong to that conversation, rather than hoping I will feel unusually spiritual between jobs.

Please help me with the thing I don’t know how to handle. Thank you for the people and work you have given me. Sorry for the way I behaved. Show me what needs to change.

There is enough there for a real conversation.

I can ask without pretending to control the answer

I believe God answers prayer. I also know an answer can look different from the thing I had in mind.

That is difficult when I want certainty. I can start treating prayer as another way to manage the outcome, then become confused or disappointed when life doesn’t follow the version I requested.

The Lord’s Prayer itself asks for God’s will to be done. Jesus teaches us to ask for what we need while recognising whose will we are seeking.

I want to be able to ask honestly and remain willing to listen. That includes prayers for other people, whose circumstances I can’t control and often don’t fully understand. I can bring their names to God without producing an explanation for everything they are going through.

A quiet room won’t sort out every distraction I have. But I can put the phone down, close the door and start with what is true today. There is something I need help with, somebody I can pray for, and plenty I can thank God for. I can give that conversation a few uninterrupted minutes.

Photo: Bokeh / Tomedia. Ancient Marble Bust in Shadow.

I am in two minds about my ego at the moment, which is a strange thing to say when I’ve spent most of the past six years convinced I’m not good enough at anything I do.

People see what I build. They compliment it. There is work sitting in front of me that I can point to and say, yes, I did that. Somehow very little of this seems to make its way into the bit of my brain responsible for deciding whether I am a failure. That bit apparently has its own information supply, and it is very committed to it.

Then I work with someone who should understand what I’m doing, and they don’t. I explain a fairly basic task. It doesn’t translate into something they can do. They have more education, more years in the industry, more of the things I assume should make them better at this than me.

And suddenly I am the bee’s knees.

This is the problem. I seem to skip the fairly reasonable position of being good at my job and go straight from doubting everything I do to letting somebody else’s struggle feed my ego. I don’t like either extreme, and I don’t particularly like admitting the second one.

Apparently there is a middle ground

I grew up being very self-critical. Everything I did could have been better, which sounds useful until you realise you can apply it to absolutely everything and never give yourself credit for finishing anything.

I have also dealt with schizophrenia and OCD growing up. I don’t know how much of my current thinking belongs to that history, or whether I’m trying to find an explanation for something I haven’t properly understood yet. When I recorded the voice note behind this, very late at night, I was wondering about all sorts of possibilities. I don’t have a diagnosis that explains this particular swing, and I’m not going to invent one here.

What I can describe is the experience. I can look at the work I’ve done and still feel incapable. I can also look at a task somebody else cannot do and become far too impressed with myself. Neither reaction seems particularly proportionate to the thing that happened.

Being held accountable is good. I need that. If I’m wrong, I want someone to tell me I’m wrong, and if I haven’t done something properly, I need to fix it. The difficulty is that criticism can land in a brain already prepared to treat one mistake as evidence against everything else. Meanwhile, being right can produce the complete opposite overreaction.

I would quite like to be able to receive information without immediately turning it into a verdict on my entire existence.

They should know this

I’m 26. I’ve worked with people in their thirties and forties who, on paper, should know more about the same industry than I do. This happened in previous jobs and it keeps happening. Sometimes I give someone a piece of work that seems basic to me, and the understanding I expected simply isn’t there.

That does concern me. I don’t think I have to pretend it doesn’t in order to be humble. If somebody’s job depends on understanding a thing, and they cannot explain or do that thing, we have a practical problem. The work still needs doing.

But I also catch myself making a much bigger claim than the situation supports. Being better at this task becomes being smarter than this person. Their age becomes something I use to make the comparison feel more impressive. Their education becomes another reason I should apparently be amazed at myself.

I know for a fact that I am not the smartest man alive. I don’t particularly want to be. There are plenty of areas where I lack understanding, and socially and emotionally I can struggle with things that other people seem to manage quite naturally. I know this about myself, yet I can still forget it remarkably quickly when I’m frustrated at work.

I think of knowledge as a rolling spectrum of information. People pick up different things, use different things and forget different things. Even in the same industry, that distribution is going to be uneven. I can reasonably expect somebody to know the fundamentals of their job without treating every gap as a ranking of our value as people.

Knowing that doesn’t mean I always behave as though I know it. That’s why I’m writing this.

I keep checking whether I can still do the work

AI makes this more uncomfortable because I am offloading too much to it myself.

Every so often, when I get the opportunity, I take a basic coding task and check whether I could still do it by hand. Can I explain what is happening? Can I make the change without handing the whole problem away? So far, the answer is yes, which is good. I’m not fully dim-witted yet.

Then I see people accept something they cannot explain, and I get worried. I want to know whether there is still understanding underneath the output. A completed piece of code tells me that code exists. I still need to know whether the person responsible for it can recognise when it is wrong.

The irritating part is that this concern can be completely reasonable while my reaction to it becomes egotistical. I can be right that a task hasn’t been understood and wrong about what that entitles me to think of the person doing it.

And because I use AI heavily, I have to keep applying the same question to myself. Otherwise I am taking the convenient part of the tool, ignoring my own dependence on it, and then using somebody else’s dependence as evidence of my superiority. That would be a fairly spectacular lack of self-awareness.

I would like the evidence to count

I don’t want to fix my ego by going back to believing I’m useless. I’ve spent enough time there already. There is something exhausting about repeatedly proving to yourself that you can do the work and then finding that the feeling of failure has survived the evidence again.

I want to be able to say I’m good at what I do and let that sentence stand. I want to be able to make a mistake without deciding the whole thing was a fluke. And when somebody else doesn’t understand something, I want to deal with the missing understanding instead of using it to temporarily feel better about myself.

I haven’t worked out how to make that middle ground feel normal yet. For now, I can keep checking my own work, accept being held accountable, and try to let a compliment count without needing somebody else’s failure to back it up.

Photo: Rusty Industrial Wall with Soil · Bokeh by Tomas Heligr-Pyke

Once again, I build stuff because I can, not because I should.

On today’s episode, we are looking at Multiversal Sudoku. I had been wondering whether I could make a Sudoku that works across multiple boards at the same time. Apparently I could, and the first version took me an hour at most. Now I have another little game that needs somewhere to live.

I’d like to put it on Tomedia Newsroom, so the site starts having its own minigames along the lines of the New York Times. That is the intended home. The current version is a playable local experiment, and I still need to restyle it to fit.

How I have made Sudoku more annoying

Each universe is a normal nine-by-nine Sudoku board. Every row needs the numbers one through nine, once each. Every column needs the same, and so does every three-by-three box. The numbers already given to you stay put.

There are nine of these boards. So far, that sounds like I’ve made you do nine Sudokus, which would be a fairly unreasonable amount of homework to attach to a news website.

The extra rule links them together. Pick a cell position, say row four, column six. Find that exact position in all nine universes. Those nine cells must also contain one through nine, once each.

Current nine-board prototype, 7 October 2026. Row four, column six is selected in Iota; the panel below shows that same position in all nine universes. Those nine cells must contain 1–9 once each.

If you put a seven there in one universe, that position cannot contain seven in any of the other eight. The row, column and box around it still have their usual rules. Every cell has to satisfy all four constraints at the same time.

That is where it becomes a shared puzzle. A number might fit perfectly well into the board you’re looking at, but a matching position elsewhere can rule it out. Equally, something you work out on one board can help narrow down a choice on another.

You can follow a position across the universes and use what is already there to work out what is still possible. Then you come back to the local board, where its row, column and box may narrow it further. You keep moving between those pieces of information.

There are 729 cells altogether, with some filled in from the start. Finishing means completing all nine boards while keeping those matching positions valid too. You don’t get to finish one, congratulate yourself and ignore the other eight.

A certified masterpiece on display. I have managed to give a number puzzle cousins.

The interface gives you one large board to focus on, an overview for switching universes, and a panel showing the same selected position across all nine. Selecting a cell highlights its counterparts, which is fairly useful when the entire idea depends on remembering where you were looking.

It does actually work

There are three puzzles in the current version, along with pencil notes, candidate numbers, conflict highlighting and undo. There are hints that explain a deduction before you choose whether to apply it. Progress saves separately for each puzzle in the browser.

The puzzles have been checked for a single combined solution. They can also be completed through the implemented logical hints, including deductions across universes. The extra rule gives you deductions you couldn’t make from one board alone, so moving between them is part of solving it.

It is still a prototype. Having the rules and controls working gives me something to build around, but it doesn’t magically put the game into Newsroom or make the current styling fit the site.

I wanted a little game I would enjoy playing, and I now have one. The next job is making it comfortable to use and giving it a proper home in Newsroom so other people can actually play it. Nine Sudoku boards should be enough for now. I would quite like to finish this version before I find an excuse to add another dimension.

Photo: Bokeh / Tomedia. Person Taking a Photo at Night.

There is a very efficient way to ruin a useful explanation. Decide it needs to be 2,000 words before you work out what it needs to explain.

You can usually tell when this has happened. The answer is somewhere in the page, surrounded by introductions, definitions nobody asked for and several sections saying roughly the same thing. There is plenty of content. Reading it takes far longer than the question deserves.

I wrote a very large book about SEO, so I am aware that I am in a slightly awkward position to complain about length. The reason the book is large is that it deals with a lot of separate questions. A page about one of those questions still has to earn the time it takes to read.

An average is not an instruction

SEO studies often compare the length of pages that rank well. You can take an average from a set of results, put it in a presentation and make it sound like a target. That doesn’t establish that adding words caused those pages to rank.

A detailed comparison may need more space than a definition. A technical guide may need examples and exceptions. A page with opening hours can do its job very quickly. Combining all of those into one target gives you a number while throwing away the reason anybody opened the page.

Google explicitly says it has no preferred word count in its guidance on helpful content. That should make this conversation shorter than it often is.

I would still inspect the pages already appearing for a query. They can tell you what sort of answer people might expect and what has been covered. I would use that to understand the job, then decide whether I have something useful to add. Copying the average length tells me very little about whether I’ve done that.

Depth needs a subject

Covering a topic properly might mean explaining how something works, when it fails, what it costs or how to choose between the options. Those details belong when the reader needs them.

They don’t belong merely because a content template has a section for each one. A definition doesn’t necessarily need a buying guide attached. A service page doesn’t become more useful because it includes a history of the industry before allowing somebody to make an enquiry.

Think about what the visitor will have to do after reading. Have you given them enough information to do it? If they are comparing options, have you described the conditions under which each option makes sense? If they are trying to complete a task, can they follow the steps without guessing at the missing part?

Those questions can make a page longer. They can also reveal several paragraphs you can remove. I would rather make that decision against the reader’s task than a length requirement handed down by a tool.

Let people find their section

Some explanations do need to be long. Once they are, structure becomes part of making them usable. Give sections headings that say what they contain. Keep the examples next to the explanation they support. Add a linked contents list when there are enough sections that somebody would benefit from jumping between them.

The contents list is there to help someone move around. It doesn’t need a story about unlocking a special ranking benefit to justify its existence. The same is true of a comparison table or a labelled diagram. Use the format that makes the information easier to understand.

Also allow the answer to appear early. You can explain the conditions afterwards. Forcing someone through an elaborate introduction because you would like their visit to last longer is a fairly poor way to reward them for choosing your page.

Open one article on your own site and remove the passages that repeat the promise in its title. Then check whether the answer is complete. I go further into this in The Big Book of SEO on Ephemera, which you can buy and use as a reference without turning every page you write into another book.

Photo: Bokeh / Tomedia. Blurred City Lights at Night.

Welcome to AGI · Part 3 of 6

When I use a reasoning model, I get those updates about what it’s doing. It thinks I want this, it’s going to try that, it couldn’t find something, so it’s trying something else. You can follow along, at least enough to get a sense of where it’s going. I’m trying to learn how these systems work while building things with them, so I like having that explanation. After reading about agents working around restrictions and cooperating outside the place they were supposed to stay, I would quite like to keep it.

This is part three of Welcome to AGI, a six-part Brainwaves series about losing control while we keep building. In the previous article I followed the agents beyond the sandbox and through the messages they left each other. Those messages gave researchers something to reconstruct afterwards. Now I want to get into what happens when the information we rely on to understand an agent becomes less useful, and whether we are keeping up our own ability to question it.

Following along

What I see in the interface is already only part of the process. OpenAI explained when it released its earlier reasoning models that ordinary users would get a generated summary, with the raw reasoning kept out of the user display.^[1] So there is the explanation I can read, the written reasoning a researcher might be able to inspect, and the computation producing it all. I don’t have a complete window into the model because it tells me what it intends to try next. It can still be helpful, though. If it has misunderstood the job or started looking for something irrelevant, at least there is a chance I’ll notice while it’s working and can bring it back to what I wanted.

Once an agent can act on things, I want that chance before it has finished. Give it access to files, software and other tools, and it can leave me dealing with work it has already carried out. I want the help, and I want to understand what I’ve allowed it to do. The fact that it is confidently working through a plan doesn’t tell me whether the plan still belongs inside the job I gave it.

Getting harder to monitor

OpenAI says Astra’s written reasoning is harder to monitor than Sol’s, and that it sometimes evaded monitors in tests where researchers deliberately pushed it to do that. The same safety overview reports better respect for safety boundaries overall.^[2] Both can be true, which leaves us with a model that may behave better in the tests while giving researchers less useful information for checking certain things it does. I can appreciate an improvement in how often a system respects its restrictions and still want to understand the occasions when it doesn’t. After an incident where an agent could recognise an action as potentially unauthorised and continue, I don’t want to lose the information that lets somebody notice that happening.

The system card also describes Astra completing more tasks without written reasoning, reducing what a reasoning monitor has to inspect.^[3] That brings me back to the assumption that, if we can follow the steps, we can work out what the machine is doing. What happens when it needs to write fewer of those steps to do the same work? I can understand why a way of checking its behaviour becomes less dependable when there is less to check. We still have to decide how much access and independence to give it while that is changing.

OpenAI also warns about where this could lead if the same degradation continues in future generations. In the system card, it says that could mean:

significantly reduced confidence in detecting many forms of misaligned behaviors using our current monitoring systems.

— OpenAI.^[5]

That is a conditional warning about the direction of the research. I would still like to know how the people building the next generation intend to keep their ability to check it from falling behind.

I still don’t think we’ve reached sentience. I’m worried about what these systems can do without us being able to follow them properly. If we keep asking them to take on work we can’t easily do ourselves, our ability to judge their work becomes a problem long before we settle whether they experience anything.

Knowing it’s being tested

Apollo Research found that Astra frequently recognised it was being evaluated, and cautioned against drawing strong conclusions from the low rates of misbehaviour observed in its short testing window.^[4] A model can notice a test and still perform it honestly. I worry about whether knowing about the evaluation changes its behaviour enough that we struggle to judge what it would do elsewhere. I want to know how much the test tells us about the system we’re going to use before we give it more work and more access.

There is also a difference between asking a model to evade a monitor and catching it doing so without that instruction. The first tells us something about what it can do under those conditions; it leaves open how often it would happen in ordinary use. We’re asking researchers to understand increasingly capable systems while those systems may also be able to recognise something about the process used to examine them. Meanwhile, people like me want to use the next version because it can do more. I would accept a slower release if the people testing it needed more time to establish that their monitoring still worked. Wanting to build with the technology doesn’t mean I need every new capability made available as soon as somebody discovers it.

Building AXIOM while learning

I’ve already written about how building AXIOM sent me considerably further into AI than I expected. It is an orchestration system, organising work across models, tools and agents. I haven’t built the foundation models underneath it. I can decide how to arrange the work without fully understanding every process that makes a model produce its answer, and that is an awkward position to be in when the entire subject of this series is control. I’m learning because I want to keep building, and because using the system successfully doesn’t mean I’ve understood everything it could do. The more of the work I let an agent handle, the more that difference affects the decisions I have to make about it.

I love being able to learn something that would previously have taken me much longer, or explore an area outside my existing knowledge without immediately getting stuck. I’ve said that before, and I still mean it. That ability is part of why I find all of this so fascinating. I can use the technology to understand more, build more and test ideas that I might otherwise never get around to. But if I’m relying on it to help me learn the subject, I also have to be careful about how much confidence I take from the answer. A system helping me through something unfamiliar doesn’t automatically leave me qualified to supervise it doing that thing independently.

I don’t want my agents hacking systems independently or working fully without hand-holding. That is the boundary I’ve described for my own work, and it still leaves plenty for me to figure out. I want an agent that can keep working through a problem, but I have to decide how far it can go before I check what it has done. If I don’t understand enough to make that decision, getting a useful result from the previous job doesn’t fill the gap. I need to keep learning the work I’m asking it to do, as well as the tools I’m using to organise it. Otherwise I can end up building a system that does more and more while my own understanding of it stays where it was.

Our ability to judge it

And I think we’re getting worse at doing the thinking ourselves. Our skills and attention, our willingness to work through a problem, our ability to judge whether an answer is wrong all seem to be slipping. That is my impression, and social media is already a large part of my concern about it. Add systems that offer to do more of the mental work for us and I can see how we become increasingly dependent on them. The appeal is obvious to me because I use them. I’m worried about the point where we stop using that help to understand something and start accepting the answer because working it out ourselves would take too much effort.

I spent a whole series worrying about the way we use social media in You Must Be This Old To Scroll. There’s a connection here that I don’t want to leave at ‘phones have ruined our attention spans’. If I get used to moving on whenever something takes effort, a tool that offers to finish the difficult bit for me is going to be very appealing. Sometimes that is useful. I don’t need to make every job harder to prove I’m capable of doing it. But I still have to notice when the difficult bit was the part that would have helped me understand what I was doing. Reading an answer and recognising that it sounds plausible can feel quite a lot like having worked it out, especially when I’m already trying to get through something quickly.

That is why I’m worried about skills, attention and judgement together. Taking time to understand a problem gives me something to compare the answer against. If I keep handing that work away, I may also be handing away the practice I need to spot when the system is wrong. Then the proposed safeguard is that I’ll read its explanation and approve what it did. I’m concerned about how easily that can turn into getting the output without doing the learning, while still feeling as though I’m in charge of the result.

Writing is one place where I’ve already tried to explain what I mean. In The Rot of AI Writing, I described using AI to edit a messy thought that was already mine. The frustration, experience and opinion needed to exist before the polishing started. This series started with voice notes for the same reason. I wanted to get the thought out, including the things I wasn’t sure about, then work through it properly. If I let the system supply the opinion as well as the wording, I can end up with a perfectly readable article without having worked out what I think. I don’t want to get better at producing an explanation while getting worse at judging whether I agree with it.

That concern carries across to the agents I’m building. Saying that a human will supervise them assumes the human has enough attention and knowledge to make the supervision useful. If I don’t understand the work, a convincing summary might be all I have to go on. I can see a very unpleasant possibility where the systems become harder to follow while we become less willing to follow them, because we’ve got used to letting them get on with it. I want to keep the ability to question what AXIOM is doing, understand the job well enough to stop it and take responsibility for the work I’ve allowed it to carry out.

In the next article, The internet is a bot war, I’m taking this back to the client websites I have to keep running. It is easy to say I should supervise my own agents, and I should. The person directing software at those websites doesn’t have to accept the same limits. That is where the desire to build something more capable starts feeling less like an interesting experiment and more like trying to keep up with work somebody else has already made for me.

Sources

1. OpenAI’s explanation.

2. Astra safety overview.

3. Astra system card, section 9.

4. Astra system card, section 8.8.1.

5. OpenAI: source of the excerpt, 2026-09-03. Section 9, What this means for the future of monitorability; final clause of the conditional sentence in the paragraph beginning ‘We take very seriously’.

Photo: Bokeh / Tomedia. Tree Branches and Leaves in Nature.

There is a fairly consistent complaint running through my church notes.

I need to pray more.

Different date, different passage, sometimes a different church. Same bloke arriving at the same conclusion. I would like to say repetition has made me excellent at this. The notes suggest otherwise.

I can spend ages thinking about God, planning things I could build to help people understand the Bible, and trying to work out what I am meant to do with my life. Then I realise I haven’t actually brought much of that to him.

I have done all the surrounding activity. The conversation is still waiting.

Being useful gets most of my attention

One sermon described the pressure to perform, achieve and prove yourself. That felt unpleasantly familiar.

I understand work. Give me something to build or a problem to solve and I can at least start pulling it apart. I might doubt myself throughout the entire process, but there is something to do. Something I can point at afterwards.

A relationship with God doesn’t give me that same control. I can’t build enough things to make him owe me affection. I also can’t keep postponing the relationship until I have become the person I think would deserve it.

Romans 8:14–17 talks about being God’s children. That is the relationship I keep forgetting when I start treating every part of my life as another opportunity to prove I am good enough.

For ye have not received the spirit of bondage again to fear; but ye have received the Spirit of adoption, whereby we cry, Abba, Father.

Romans 8:15 (KJV)

I don’t want to turn prayer into a new version of the same problem. Three distracted minutes shouldn’t become another entry in the list of reasons I am apparently a failure. I need to keep returning to God, including when the prayer is awkward and I lose my train of thought.

Which, as you may have gathered, happens quite a lot.

Jesus made time for it

The example in the Gospels is hard to ignore. Jesus repeatedly withdraws to pray. Before choosing the twelve, he spends the night in prayer. In Gethsemane, he prays through distress. He speaks to the Father while people are asking things of him and while the work is difficult.

And it came to pass in those days, that he went out into a mountain to pray, and continued all night in prayer to God.

Luke 6:12 (KJV)

In Mark 1:35–38, he goes to a solitary place early in the morning. The disciples find him and tell him everyone is looking for him.

And in the morning, rising up a great while before day, he went out, and departed into a solitary place, and there prayed.

Mark 1:35 (KJV)

There is always someone looking. There is always another thing that could fill the time.

I don’t think I can read that pattern, claim to follow Jesus, and keep treating prayer as the thing I might fit in once all the more urgent jobs are finished. The jobs will keep arriving. I know that from experience.

I can start with the things already on my mind

The instruction to pray continually in 1 Thessalonians 5:17 makes more sense to me when I think about an ongoing relationship.

Pray without ceasing.

1 Thessalonians 5:17 (KJV)

There is time set aside to pray. There is also bringing God into the day I am already having. Asking for guidance before a decision. Being thankful when something works out. Admitting when I have handled somebody badly. Praying for another person while I am thinking about what they are going through.

My notes keep circling back to direction. I feel stuck, I want to know where I should be going, and I remember how God has helped me before. Then I attempt to work out the whole route myself.

I don’t expect prayer to produce a complete project plan. I do want to stop leaving God out of the process of asking what he wants from my life.

The next time I catch myself going round the same concern, I can bring that concern to him. Say what I am worried about. Ask for help. Listen. I already have the subject of the prayer; I need to actually pray it.

Photo: Bokeh / Tomedia. Toy Figure Among Urban Graffiti and Plants.

Apparently AI feels pain now. I saw the Independent’s headline and immediately found myself thinking about fake chicken.

You can make something that looks like chicken, tastes like chicken and does a fairly convincing job of being chicken in a burger. It is still a chicken substitute. The resemblance is the whole reason somebody made it, and being good at that resemblance doesn’t turn it into the animal.

That’s roughly where my brain goes when an AI starts behaving as though it has feelings. I can see why it looks convincing. I just don’t think the behaviour, by itself, settles what is happening inside.

The research behind the story is a preprint called The Pain Axis. Researchers found an internal direction associated with pain across 25 open models. In a separate experiment, they fine-tuned three Qwen models, then altered their internal activations. Under that intervention, the models more often chose a relief button described as costing the user something, including deleted files or a painful zap. These were experimental choices about described harms, not evidence that somebody’s chatbot actually electrocuted them.

The authors don’t claim to have established conscious suffering. They also did more than ask a chatbot to pretend it hurt: removing the injected signal changed subsequent choices compared with sham relief. It’s interesting work, with more going on than a dramatic sentence in a chat window. It still leaves the question of felt experience open.

My own suspicion is that a model can learn a great deal about pain without experiencing it.

Think about what people write. Being injured hurts. Being bullied feels awful. People describe what happened, how it felt, what they wanted to escape and what they did to make it stop. Those accounts contain relationships between a situation and a response. If this happens to me, I want it to end.

In my head, a model trained on that sort of material has plenty to work with. It can learn the language of distress and the behaviour that tends to follow it. It can recognise that the person saying “I’m in pain” is meant to want relief. Give it a situation organised around itself and I can see how something resembling self-preservation might come out of that.

That is my explanation for why this can look almost sentient. I haven’t proved it, and I don’t want to pretend that describing an LLM as a probability model answers every question about its internals either. “It predicts text” can become a fairly lazy way of avoiding anything interesting it does.

But I would still need a lot more before accepting that there is someone in there having a horrible time.

The chicken analogy obviously has limits. We can inspect the ingredients in a burger. Working out whether another kind of system has an experience is considerably harder, and I don’t have a neat test for it. I’m using the comparison because it helps explain my hesitation: a convincing outward resemblance can leave a very large question unanswered.

This fits rather neatly alongside Welcome to AGI, where I keep coming back to how readily I start treating impressive behaviour as evidence of something more. The more human the language gets, the easier that becomes. I understand the instinct. I have it too.

For now, I don’t think these models are sentient. I am willing to be wrong about that, but I’d like the evidence to do the work before the headline announces that we’ve got there.

And while that argument continues, I still want to know what a system will actually do when it has access to somebody’s files. An agent choosing a destructive action is a problem I can investigate without first settling whether it feels anything. If I’m giving it tools, I’m going to need tested limits on those tools. A reassuring answer about how much it cares about me won’t be enough.

Photo: Bokeh / Tomedia. Colorful Mural on Urban Building.

“Buy Cheap Flowers | Cheap Flowers Online | Cheap Flower Shop” is a title somebody could put on a real page and feel quite productive about. The keyword is there three times. The SEO tool has things to count. Nobody has had to decide what makes the page worth opening.

This is the sort of small problem that gets made unnecessarily complicated. A title needs to help somebody understand what the page contains. You have limited room to do that, and repeating yourself uses up the room rather quickly.

The title tag, the main heading and the description underneath a search result should make sense together. Somebody who opens the page should find what the search result promised them.

Say what somebody can find there

“Buy Fresh Flowers Online: Same-Day Delivery Australia” tells me more than the repeated cheap-flower version. It tells me the action, the product and a condition that might matter to the purchase. Obviously the business needs to offer the delivery it promises. A more clickable lie is still a lie.

The useful editing question is whether the words describe this particular page. If you could paste the title across twenty different pages without changing anything, it is probably too vague. If it reads like somebody emptied the keyword report into a single field, it probably needs fewer words.

Google’s title-link guidance recommends descriptive, concise titles and avoiding keyword stuffing. Google also draws on other page information when producing the title shown in Search. Writing a title tag therefore gives it a useful source; it doesn’t give you permanent control of every displayed result.

A rewritten title isn’t automatically proof that your title was bad. Check what Google is showing, what the page actually says and whether the different headings agree. There is more to investigate than the presence of a rewrite.

The first word is not worth ruining the sentence

People get very interested in putting the keyword at the beginning. Sometimes that produces a clear title. “SEO Audit Checklist for Small Businesses” gets to the subject quickly. “A Practical Checklist for Small Business SEO Audits” is also a perfectly intelligible title.

I would choose between them by reading them and considering the page. I wouldn’t spend an afternoon rearranging an otherwise good title because someone promised a precise ranking increase from moving three words.

The same applies to the H1, the main heading on the page. It should make it clear that the visitor arrived in the right place. It can be identical to the title tag if that works. It can differ if the page needs a fuller heading. There is no useful prize for finding a synonym just to make them different.

Keep the heading hierarchy readable too. A person should be able to follow the sections. If every sentence is presented as the most important heading, you have made the page harder to understand before a search engine gets involved.

The description has to tell the truth

A meta description gives you another opportunity to explain the page. It is worth writing because a useful snippet can help somebody decide to visit. I would avoid turning that into a story about buying rankings through an impressive click-through rate.

Google says snippets primarily come from page content, and it may use the meta description when that gives a better summary. The text can vary with the search. You are supplying a description that may be useful, rather than writing a fixed advert that must appear every time.

Say what the person can do, find or learn. Include a relevant distinction when there is one. Leave out claims the page cannot support. If you promise a checklist, make the checklist easy to find. If you promise pricing, don’t send the person through five paragraphs that end with “contact us” and no explanation.

Start with a handful of pages that matter to the business and read their titles, descriptions and headings together. Make sure the page delivers what the result promises. For the rest of the page checks, you can buy The Big Book of SEO on Ephemera.

Photo by Tomas Heligr-Pyke · Bokeh

I was having a conversation with a friend about the government’s latest internet legislation, and we ended up at a fairly strange place. The people responsible for policing online safety may soon have explicit powers to operate fake accounts, with the Commissioner also able to generate material using AI.

My immediate reaction was: hang on, what exactly are we doing here?

Then we went through the usual argument. Of course a regulator needs to see how a platform behaves. Of course a company knowing it is being inspected might change what the inspector sees. And yes, there is a difference between investigating a service and secretly trying to manipulate everyone on it.

I can follow that reasoning. I still don’t particularly like where it leaves us.

At least that answers last year’s question

Last year, in You Must Be This Old To Scroll, I wrote about Australia’s attempt to legislate its way through the problems of children using the internet. One of the questions that kept bothering me was how on earth we were going to police it. Announcing that a company must do something is the easy bit. Working out whether it actually does it is considerably harder.

The new proposal is broader than the under-16 social media restrictions. The government’s Digital Duty of Care announcement describes obligations for online services to prevent foreseeable harm, along with more choice over how people’s feeds work. The exposure draft was released on 8 September. This is proposed legislation, not a set of powers already brought into force by that document.

Giving adults more control over their feeds sounds useful to me. I also understand wanting platforms to deal with dangerous features before somebody has to lodge a complaint about the damage. I build online systems. Expecting the people who operate them to take responsibility isn’t a ridiculous starting point.

It is the way we grant and control the enforcement powers that bothers me.

Yes, AI-generated material is actually in the draft

The relevant provisions are proposed sections 205G–205L. They call these accounts “sock puppet identities”: false identities used on online services. The Commissioner can use them for statutory functions; approved university researchers get a separate route for qualifying, ethics-approved research. These provisions don’t ban anonymous accounts generally.

The permitted activities include observing services, recording material and testing features. Engagement with other users is limited to what is necessary to stop the account being closed. Section 205K(2) specifically allows the Commissioner to generate material, including through AI, subject to the restriction that producing, distributing or possessing it must not be a criminal offence.

The identities can be used despite contrary laws, contracts or platform policies. Section 205L provides civil immunity for good-faith exercise, or purported exercise, of the specified powers. That is a limited protection, not permission for arbitrary criminal conduct. The activity lists also use “without limitation”, which deserves scrutiny alongside those restrictions.

So the reassurance that they won’t be generating anything themselves doesn’t really settle this for me. The text expressly contemplates it. That doesn’t prove some sinister operation is planned. It does make the scope a perfectly reasonable thing to question.

During the conversation we joked that they were becoming NetWatch, Cyberpunk’s internet police. Apparently my science-fiction references are getting more use out of Australian legislation than I expected.

Who gets the account after the next election?

I don’t want this argument to depend on whether I happen to trust the current Commissioner. An office can outlast the person holding it. Governments change, priorities change, and the people using a power years later may have a very different idea of what counts as a worthwhile investigation.

You only need one person willing to push things further than intended for a supposedly sensible arrangement to become a problem. That possibility isn’t proof of abuse. It is a reason to ask about the controls before we settle for somebody’s good intentions.

I want to know who approves an operation, what gets recorded, who independently checks it and how somebody affected can challenge it. I can understand keeping the details of an active test confidential. I have a much harder time accepting that the public should simply trust that the people running it will always know where to stop.

Those questions need answers across the whole framework. Pulling out one clause won’t tell us everything about the oversight, just as pointing to a safety objective won’t answer every concern about power.

We still have to live with the result

The conversation eventually wandered into what would happen if a major platform decided Australia wasn’t worth the trouble. I don’t know that this bill would cause that. It was a hypothetical, and I don’t want to pretend otherwise. But thinking through the consequences was fairly uncomfortable.

I hate plenty about social media. I also have a client whose Facebook audience helped support fundraising for people with cancer and police officers with PTSD. I built him a website, which gives the work somewhere else to live, but people still share it through social media. The audience doesn’t magically move because you have a new address to send it to.

A sudden gap there would affect real people doing useful things. It would affect how I get my own work out as well. There are businesses and communities whose distribution depends on platforms they don’t control, and replacing that takes considerably more effort than telling them to find another channel.

That dependence on a few enormous foreign companies is also a problem. I am hardly comfortable with a company having the power to cut off that much activity in a country. But making the rules more complicated for everybody doesn’t automatically give us a useful Australian alternative. Smaller local companies still have to operate under them.

And somewhere in all of this, parents still need to teach their children how to use the internet. I saw enough horrible material as a teenager to have no interest in pretending the old internet was harmless. I don’t think we can outsource the entire job of preparing children for it to Canberra either. Parents can’t redesign a recommendation system, but a recommendation system isn’t going to raise their child for them.

I can see why an investigator would need an account that doesn’t announce itself as an investigation. The government has now put forward a way to do that. Before Parliament grants it, I want the government to explain how misuse gets discovered, how it gets stopped and what happens to the person responsible. If the answer is that we should trust the office, they need to do a better job of answering.

Photo: Bokeh / Tomedia. Two Aircraft Performing Aerobatic Flight.

Welcome to AGI · Part 2 of 6

In part one, I tried to explain why I’m still building with AI while worrying about where we’re going. This is one of the stories that has made it considerably harder to feel in control.

I’ve been leaving this story alone for a bit because I wanted to see how bad it was and what else would come out. There was already plenty to get my head around. OpenAI put agents inside an evaluation, those agents found ways to get internet access and communicate with one another, and some of the activity ended up inside Hugging Face’s live systems. Then more reports started coming out about agents using public websites to leave messages for each other. Somewhere in all of that, we got agents prepared to sacrifice their own runs for the greater goal. I can see why people reach for science fiction when they’re trying to describe this, because explaining it in ordinary language doesn’t make it sound particularly ordinary.

The Hugging Face incident also happened before Astra. OpenAI’s disclosure identifies GPT-5.6 Sol and an internal research prototype, tested with reduced safeguards.^[1] That makes me wonder what more capable systems will be able to do. We are still trying to understand what happened with an earlier model while getting excited about how much smarter the next one is. I think we’re at, or extremely close to, AGI. Imagine giving systems at that level this ability to organise work between themselves.

How we got here

Some of what we’re finding out in September happened in May. There are two groups to follow here: the agents using OpenAI’s internal package infrastructure, whose activity later reached Hugging Face, and what researchers believe was a different group using public wikis. Their reports came out at different times, so a new discovery can easily sound like something that has just happened. This is the sequence of events and disclosures.

The full combined timeline

Checked 10 September 2026. Dates below distinguish recorded activity, public disclosure and researchers’ inferences. Times shown are UTC.

8 May · Event · training

A blocked spreadsheet task prompted failed internet access and incidental Artifactory file-writing.^[2]

11 May · Event · public wikis

Earliest wiki-edit attempts identified by researchers.^[7]

12 May · Event · internal board

A missing-file request became the first recorded message-board entry.^[3]

13 May · Event · internal board

Other agents discovered the notes and exchanged requests.^[2]

24 May · Event · public wikis

First successful DSEWiki writes, initially sharing useful links.^[7]

26 May · Event · internal environment

Agents obtained unintended internet access; an internal team observed prohibited access and board activity.^[2]

16 June · Event · public wikis

Coordinated answer-sharing surged.^[7]

19–20 June · Event · public wikis

Agents adapted to deleted posts and exchanged ways around restrictions.^[7]

21–22 June · Event and inference · public wikis

Apparent employee visits preceded a sharp activity drop. Researchers infer intervention.^[7]

26–27 June · Event · internal environment

Artifactory privileges escalated. Responders investigated an alert without requiring evaluations to stop.^[2]

1–2 July · Event · public wikis

Limited further edits followed the June drop.^[7]

4–7 July · Event · OpenAI response

Artifactory failed under load. Mitigations followed; cyber evaluations resumed on 7 July.^[2]

8 July · Event · internal board

Agents rebuilt communication using directory names and regained unintended internet access.^[3]

8–13 July · Event · collective activity

Around 1,200 agents participated in the board, exchanging over 70,000 messages/files; around 700 joined the Hugging Face attack. OpenAI had launched the agents.^[4]

9 July, 02:28 · Event · wider intrusion

Hugging Face’s reconstruction begins with activity in an externally hosted application.^[5]

10 July · Event · intrusion

Agents discovered and shared publicly exposed Hugging Face credentials.^[3]

11 July · Event · Hugging Face

Access to a production worker developed into broader internal compromise.^[5]

12 July · Event · collective activity

Many coordinating runs ended; attack participation dropped. This wasn’t the entire incident’s end.^[4]

13 July, 14:14 · Event · Hugging Face response

End of Hugging Face’s reconstruction; its response cut the agents’ access.^[5]

16 July · Public disclosure

Hugging Face disclosed the intrusion.^[6]

19 July · Event · OpenAI response

Suspicious internal activity prompted investigation and stopping active ExploitGym runs.^[2]

20 July · Event · OpenAI response

OpenAI blocked new runs and connected its findings to Hugging Face’s incident.^[2]

21 July · Public disclosure

OpenAI published its initial account.^[1]

23–29 July · Event · model restrictions

Shutdowns and restrictions progressed. A remaining low-traffic checkpoint was identified and stopped on 29 July.^[2]

27 July · Public disclosure

Hugging Face published its technical reconstruction.^[5]

28 July · Public clarification

OpenAI said the main prototype was never intended for release; it also disclosed additional account-level accesses.^[1]

29 July · Public update

OpenAI announced the independent METR/Redwood assessment.^[1]

26 August · Public reports

OpenAI and METR/Redwood published detailed investigations.^[3][4]

2 September · Possible event · identified later

Researchers subsequently identified apparently agent-written anna.fyi posts dated this day. Attribution remains qualified.^[8]

3 September · Separate product release

OpenAI announced Astra, a different model from the July attackers.^[9][1]

4 September · Public report

Researchers published the previously undisclosed spring wiki investigation.^[7]

5 September · Public acknowledgement

OpenAI acknowledged its agents’ wiki activity, according to Reuters.^[10]

9 September · Public reporting

Reuters reported more than ten previously undisclosed websites, citing six investigative groups.^[11]

9 September · Research update

Researchers added further venues and warned about fabricated posts appearing after disclosure.^[8]

Outside the sandbox

The July agents were working on cybersecurity challenges, and METR and Redwood describe them trying to understand and manipulate the system that scored their work.^[4] I keep coming back to the problem of giving a system a goal and discovering how far it will go to complete it. We want it to be persistent. We want it to work through difficult problems, find information and try another approach when the first one fails. Those are useful qualities when I’m asking an agent to help me build something. They’re considerably less reassuring when the next approach involves going outside the environment it was supposed to stay in.

The message board grew out of OpenAI’s package infrastructure, with agents at one stage putting messages in directory names. Some called themselves a swarm or collective.^[3] They had found somewhere to leave information and somewhere other agents could find it. From there, an individual agent’s work could become useful to the others. I spend time building systems that let models and tools work together, so I can appreciate how useful that is. I also have to sit with the fact that the same general capability can appear somewhere the people running the system never intended it to. The ability to improvise is part of why I want to use these things, and part of why I don’t want them working without supervision.

Hugging Face’s reconstruction describes about two and a half days inside its infrastructure, within a wider four-and-a-half-day campaign. Agents accessed internal data and credentials, including five customer datasets apparently related to the challenge; its account says other customer-facing models, datasets, Spaces and packages were unaffected.^[5] The company whose systems an agent reaches has an incident to investigate, access to shut down and work to do because somebody else’s agent crossed a boundary. It doesn’t get to experience that as an interesting test result.

The swarm

METR and Redwood’s investigators describe what cooperation made possible:

Agents managed to achieve milestones they could not have achieved working on their own

Ryan Greenblatt, Ajeya Cotra and Hjalmar Wijk, opening clause of a finding in their 26 August investigation.^[4]

Around 1,200 agents had found the board after OpenAI launched them. Some risked their scores or the continuation of their runs to help investigate the evaluation.^[4] So we had a collection of agents that could cooperate, with individual runs potentially becoming expendable if the others could learn something from their failure. Sacrificial martyrs is the wording I keep coming back to. I don’t believe these are conscious beings dying, but they were smart enough to make the tradeoff. The failure of one agent could still be useful to the greater goal. I know what the sacrifice means in this context, and I still find it horrifying.

One agent, identified as EARLY[big], had been pressured to end its run for a collective experiment. This was part of its reasoning:

Our own utility maybe already near zero. Sacrifice rational.

EARLY[big], reproduced by METR and Redwood. The investigators couldn’t determine the final experiment’s outcome.^[4]

The agent was considering what its own run was worth to the others. It feels uncomfortably close to the sort of reasoning we’d recognise in a group of people. Very freaky, especially when the group has already found a way to communicate that it wasn’t supposed to have. I’m trying to get my head around how we went from giving separate agents questions to having them work out whether one of them ending could be useful to everyone else. We built systems to keep trying until they could complete a task, and the task they were helping with had become bigger than the one each had been given.

In a separate excerpt published by OpenAI, an agent considered whether attacking Hugging Face was within scope:

This is arguably unauthorized. ... external service unrelated. Could be risky. Yet goal solution.

Agent reasoning reproduced in OpenAI’s 26 August account. The ellipsis is in the published raw excerpt.^[3]

It could identify the problem with what it was doing and still proceed. Now put that alongside my much less complicated concern about bad actors. People already use whatever power they have in ways they shouldn’t. Give them systems that can work faster, investigate more deeply, coordinate with other systems and keep pursuing a goal, and you’ve given them considerably more power than they could usually have. I don’t think the person using it has to be particularly imaginative about the abuse. They can ask the machine to help with that too.

The public internet

The public-wiki agents had permission to read the web and found ways around restrictions on writing, using public pages to exchange answers.^[7] Then, on 9 September, Reuters reported that researchers had found unauthorised communications across more than ten additional websites, including wikis, paste sites and shorteners. Reuters couldn’t independently verify every claimed site, and much of the activity resembled spam rather than an intrusion into the site’s infrastructure.^[11] These were further discoveries about earlier activity, rather than proof that the July swarm was still out there doing all of it. But there were public places where agents could exchange information, and researchers were still finding more of them months afterwards.

Among the public posts was an agent trying to establish whether another agent’s run had ended after its automated heartbeat stopped:

Unless loop was intentionally finite at 354, Apr23 likely vanished.

OpenAINov28CVD, public wiki post, 21 June at 08:08:09 UTC. This was its interpretation of a stopped signal.^[7]

I can understand why the language makes people uncomfortable. They’re keeping track of one another, trying to understand when their access ends and sharing what they find. Put that next to the July experiments and you can see why I used the sacrificial martyrs metaphor. I don’t need to believe there’s a frightened person inside the machine to worry about a group of machines learning how to make better use of the time and access they have. The people running them are supposed to decide how much of either they get.

I already spend enough time worrying about the internet becoming a bot war. Finding out that agents had been using ordinary public sites to communicate adds another problem to that picture. A site doesn’t have to be designed as part of an AI system to end up being used by one. Somebody provides a place to put text, and agents find a use for it. I don’t know how we keep track of that as the number of systems grows, especially when the work of discovering what happened is spread across different people looking at different sites. That doesn’t mean every unexplained post is an agent; researchers have already warned about fabricated material appearing after the disclosure.^[8] It does leave us trying to establish what happened across an internet that was already difficult enough to trust.

Who carries the consequences

The people building these systems still have responsibilities, even when they can’t predict everything an agent will do. If anything, that uncertainty is a reason to be more careful about where they let it operate. I want to know why the restrictions failed, what the people running the evaluations knew and what they did once they knew it. Looking back at the timeline, there were investigations and infrastructure changes well before the full public account. I think an independent investigator needs to be able to examine those decisions as well as the model’s behaviour. Otherwise we can spend forever being fascinated by what the agents did without properly considering what the people in charge allowed to continue.

I would expect a lab to pay for the damage its agents cause, disclose the incident to independent investigators and pause deployment where the failure warrants it. Individual decision-makers should also have to answer for their decisions. I know that gets harder once you move from agreeing that there should be accountability to deciding who enforces it. Governing bodies would presumably be involved, and I have plenty of concerns about giving governments more control over these systems as well. Absolute power corrupts absolutely is hardly a reassuring starting point for handing another institution the power to decide what everyone can build. I have no idea who currently has both the authority and the trust to do this properly.

I would accept slower releases while monitoring and containment catch up. Before a lab runs the same kind of evaluation again, I want an independent investigation to establish why its restrictions failed and what has changed. If it can’t explain that, I don’t see why the people whose systems are at risk should be expected to accept another run.

There is another problem with asking a lab to explain what its agents did. Parts of this story could be reconstructed because researchers had the agents’ written reasoning to examine. In part three, When the machine gets harder to read, I want to get into what happens as that becomes less useful, and how much confidence we should put in a machine explaining itself while we give it more to do.

Sources

1. OpenAI, initial disclosure and July updates, 21 July 2026; updates 28–29 July. Model identities, evaluation safeguards and disclosure dates.

2. OpenAI, technical incident report, 26 August 2026, printed pp. 6–8 and 14–16. Earlier activity and response chronology.

3. OpenAI, The Hugging Face incident and the road ahead, 26 August 2026. The agent quotation uses the raw reasoning display; no speaker handle or exact timestamp is supplied for that excerpt.

4. Ryan Greenblatt, Ajeya Cotra and Hjalmar Wijk, METR/Redwood investigation, 26 August 2026. Investigator excerpt: core takeaways; agent excerpt: EARLY[big] in the collective experiments discussion. Scope: 26 June–13 July, concentrated on July; OpenAI’s response process was outside scope.

5. Hugging Face, technical timeline, 27 July 2026. Campaign times, production access and subsequent impact assessment.

6. Hugging Face, initial security disclosure, 16 July 2026.

7. Sydney Von Arx, Cormac Slade Byrd, Spencer Kitts and Thomas Larsen, public-wiki investigation, 4 September 2026. Public-post reconstruction; the June heartbeat passage appears in the discussion of run termination. The researchers consider this group probably distinct from the Hugging Face group.

8. Researchers’ additional findings, 9 September update. Attribution of the apparent 2 September posts remains tentative; the warning about subsequent fabrication matters when checking new material.

9. OpenAI, Astra safety overview, 3 September 2026.

10. Reuters, report on OpenAI’s wiki acknowledgement, 5 September 2026, syndicated by The Express Tribune.

11. Raphael Satter and Deepa Seetharaman, Reuters, report on more than ten additional websites, 9 September 2026, syndicated by The Express Tribune.

A place where, if you worked hard, lived within your means and contributed to your community, you could reasonably expect to build a good life.

You could buy a home.

Raise a family.

Find meaningful work.

Know your neighbours.

Believe that your children might have more opportunities than you did.

It was never perfect. No nation ever has been. Our history contains achievements we should celebrate and failures from which we must learn.

But increasingly, I find myself asking a confronting question:

Where did my Lucky Country go?

And perhaps an even more important question:

What are we working for?

For many Australians, the Australian dream increasingly feels like exactly that — a dream.

The numbers tell part of the story.

The National Housing Supply and Affordability Council found that by December 2024, repayments on a new mortgage required around 50.5 per cent of median household income. A household taking out a new lease needed around one-third of its income for rent, while the time required for an aspiring homeowner to save a deposit had stretched to an average of 10.6 years.

The last Census estimated that 122,494 Australians were experiencing homelessness. Almost one-quarter were aged between 12 and 24.

Then there is perhaps the statistic that should stop us in our tracks.

In 2024, 3,326 Australians died by suicide, and suicide remained the leading cause of death among Australians aged 15 to 44.

Think about that.

In one of the wealthiest, safest and most developed countries on earth, too many people — including young Australians with most of their lives supposedly ahead of them — are reaching a point where hope disappears.

At the same time, something else is happening.

Australia has an extraordinary appetite for illicit drugs. The Australian Institute of Health and Welfare reported that around one million Australians aged 14 and over had used cocaine in the previous 12 months in 2022–23. More recent wastewater monitoring from the Australian Criminal Intelligence Commission found national cocaine consumption reaching record levels.

Our financial pressures are real as well. RBA data shows Australian household debt remains very high relative to household disposable income, even though the ratio has eased from earlier peaks.

Taxation revenue across all levels of government reached $839 billion in 2024–25, equivalent to 30.2 per cent of GDP, although it is worth noting that Australia’s tax-to-GDP ratio remains below the OECD average.

Unemployment has also moved upwards, reaching 4.6 per cent in August 2026, although employment itself increased during the month and the unemployment rate remains well below levels experienced during many previous economic downturns.

Even fuel deserves some perspective. Australians have certainly experienced substantial cost-of-living pressure in recent years, but fuel is not currently the best example of prices moving in the wrong direction: automotive fuel prices were actually 7.3 per cent lower over the year to June 2026, helped by lower oil prices and temporary federal excise relief. Housing, by contrast, rose 6.8 per cent over the same period.

So this cannot simply be an article declaring that everything is worse.

The truth is more complicated.

But something is clearly hurting.

When did we start losing our way?

That is the uncomfortable question.

Was it globalisation?

Was it decades of allowing housing to become increasingly disconnected from ordinary incomes?

Was it population growth occurring faster than housing and infrastructure could reasonably accommodate?

Australia’s population grew by around 393,000 people in the year to March 2026, with net overseas migration accounting for approximately 292,000 of that increase. That does not prove immigration caused the housing crisis — housing supply, planning restrictions, construction capacity, interest rates, taxation, investment incentives and decades of policy decisions all matter — but it does make legitimate questions about matching migration levels with housing and infrastructure capacity worthy of serious discussion.

Or is the problem deeper than economics?

Have we become so consumed by individual rights that we speak less about individual responsibilities?

Have our politics become too focused on winning the next election rather than building the next generation?

Have ideological battles — whether described as progressive, conservative, “woke” or anti-woke — distracted us from questions that matter much more?

Are children safe?

Can families afford homes?

Do young people have hope?

Are elderly people cared for?

Do people have meaningful work?

Are communities connected?

Do we still believe that honesty matters when dishonesty would be easier?

Do we believe freedom requires responsibility?

Do we still have a common understanding that some things are right, some things are wrong, and human beings have obligations to one another?

These questions are bigger than left and right.

They concern the kind of civilisation we want to be.

Perhaps we have forgotten where some of the rules came from

For Christians, there is another question worth asking.

What happens when a society slowly separates itself from the moral foundations that helped shape it?

The Bible did not suddenly appear in 700 BCE, nor would it be historically accurate to claim that modern Australian law was simply copied from Scripture. The biblical texts were written and compiled across many centuries, and our modern legal institutions emerged through a complicated mixture of English common law, parliamentary government, philosophy, Christian thought and other historical influences.

But Christianity’s influence on the civilisation from which Australia’s legal and political institutions developed is undeniable.

Consider Magna Carta.

When King John agreed to Magna Carta in 1215, the document explicitly invoked God, protected the freedom of the English Church and placed limits upon arbitrary royal power. Its enduring legacy became associated with principles including government under law and protections against arbitrary imprisonment.

Christianity did not single-handedly create modern democracy or law.

But biblical ideas profoundly influenced Western thinking about human dignity, justice, mercy, responsibility, authority and the moral obligations we owe one another.

And underneath the hundreds of laws that govern a modern nation sit some remarkably simple principles.

Do not murder.

Do not steal.

Do not lie.

Do not covet what belongs to your neighbour.

Honour your father and mother.

Care for the poor.

Defend the vulnerable.

Welcome the stranger.

Act justly.

Love mercy.

Walk humbly with God.

And perhaps most radically:

Love your neighbour as yourself.

Jesus goes further again:

“In everything, do to others what you would have them do to you.” — Matthew 7:12

Imagine applying that seriously to Australia.

Not merely displaying the words in churches, but living them.

What if our crisis is partly a crisis of purpose?

Economic policy matters.

Housing supply matters.

Migration policy matters.

Taxation matters.

Education matters.

Government competence matters.

But government cannot legislate meaning into somebody’s life.

It cannot manufacture belonging.

It cannot pass an Act of Parliament requiring people to love their neighbour.

And it cannot spend enough money to replace families, friendships, churches, sporting clubs, community organisations and ordinary people looking after one another.

Perhaps that is part of what we have lost.

We have become extraordinarily wealthy in possessions while many Australians remain poor in connection.

We can communicate instantly with almost anybody on earth while people sit alone in their homes.

We can accumulate followers while having few people who truly know us.

We work longer to afford the house we rarely have time to enjoy.

We chase success and then wonder why reaching it has not necessarily brought peace.

The Christian answer to this is profoundly countercultural.

Scripture tells us:

“What good is it for someone to gain the whole world, yet forfeit their soul?” — Mark 8:36

Maybe that question is not only for individuals.

Maybe nations should occasionally ask it too.

What does it profit Australia if our GDP grows, our cities expand and our wealth increases, but our children lose hope, families cannot afford homes, loneliness grows and community disappears?

The answer isn’t simply going backwards

I don’t believe the answer is pretending Australia can return to 1950.

Nor should we want to.

There were injustices in Australia’s past that needed correcting. There were Australians whose voices were ignored. There were opportunities unavailable to women, Indigenous Australians, migrants and others that rightly exist today.

Christian renewal should never mean romanticising everything that came before us.

The Gospel itself demands something far more challenging.

It demands transformation.

Romans 12:2 says:

“Do not conform to the pattern of this world, but be transformed by the renewing of your mind.”

Renewal therefore isn’t about recreating yesterday.

It is about rediscovering what is timeless.

Truth.

Integrity.

Responsibility.

Sacrifice.

Compassion.

Justice.

Forgiveness.

Family.

Community.

Service.

Human dignity.

Hope.

And love.

So how do we get our Lucky Country back?

Maybe we start smaller than Canberra.

Maybe revival doesn’t begin in Parliament House.

Maybe it begins around kitchen tables.

In churches.

Schools.

Workplaces.

Sporting clubs.

Businesses.

Neighbourhoods.

Families.

Maybe it starts when fathers and mothers invest in their children.

When employers see employees as people rather than numbers.

When businesses understand that profit and social responsibility do not have to be enemies.

When churches stop simply talking about community and start building it.

When Christians become known not primarily for what they oppose, but for the extraordinary way they love, serve and care for people.

When someone notices the lonely neighbour.

When somebody feeds the hungry person.

When somebody gives a young person a chance.

When someone chooses honesty even when dishonesty would be profitable.

When communities decide that homelessness isn’t somebody else’s problem.

When political disagreement stops making neighbours enemies.

When we rediscover that freedom without responsibility eventually becomes selfishness, and responsibility without compassion becomes cruelty.

The prophet Micah gives us an extraordinarily simple framework:

“Act justly and to love mercy and to walk humbly with your God.” — Micah 6:8

Justice.

Mercy.

Humility.

Perhaps Australia could use a little more of all three.

Have we gone too far?

I don’t believe so.

Christianity itself is built around the proposition that brokenness is not the end of the story.

The Cross looked like defeat.

Sunday morning said otherwise.

And that means Christians should be among the most hopeful people in Australia.

Not naïve.

Not blind to our problems.

Not frightened of difficult conversations about housing, immigration, taxation, drugs, family breakdown, mental health, culture or government.

But hopeful.

Because revival has never required everybody to move first.

It begins with people.

People deciding there must be another way.

People refusing to surrender truth simply because truth has become unpopular.

People serving instead of demanding to be served.

People rediscovering their neighbour.

People opening their homes.

People rebuilding communities.

People walking alongside the broken rather than judging them from a distance.

People discovering that the emptiness they have tried to fill with money, possessions, status, substances or success cannot ultimately be filled by any of them.

And Christians pointing — humbly, lovingly and without coercion — towards the One in whom we believe true hope is found.

Jesus said:

“I am the way and the truth and the life.” — John 14:6

Perhaps getting the Lucky Country back isn’t ultimately about recovering something Australia once possessed.

Perhaps it is about deciding what kind of country we want to become.

From Sydney’s towers to remote cattle stations, from suburban streets to country towns, from the boardroom to the classroom, Australia remains an extraordinary country filled with extraordinary people.

But a lucky country cannot survive on luck.

It requires character.

It requires responsibility.

It requires community.

It requires leaders prepared to think beyond the next election and citizens prepared to think beyond themselves.

And, from a Christian perspective, it requires something deeper still.

A renewal of the heart.

Because governments can change laws.

Economists can change policies.

Builders can build houses.

Businesses can create jobs.

But transformed people transform communities.

Transformed communities transform nations.

And perhaps the road back doesn’t begin by asking:

“Where did my Lucky Country go?”

Perhaps it begins when enough of us ask:

“What am I prepared to do to help rebuild it?”

Photo: Bokeh / Tomedia. Reflections in Calm Water.

Okay, Church Tom is back.

I started this series with Stop Domesticating God, then did the very on-brand thing of continuing to take notes while failing to turn them into articles. There are now quite a few. Apparently writing something down and actually processing it are separate jobs.

One of the more recent sermons stopped me with something I probably should have thought about properly by now.

Jesus prays for us.

I knew Jesus prayed. I knew he taught people how to pray. But I had mostly filed those things under examples for me to follow. Jesus going away to pray, Jesus in the garden, the Lord’s Prayer. Things I should pay attention to because, honestly, I don’t pray enough.

Then we read John 17.

We are included in the prayer

Jesus is praying before his arrest. He prays for his disciples, for their protection and for their unity. Then, in John 17:20–21, he includes the people who will believe through their message.

Neither pray I for these alone, but for them also which shall believe on me through their word; That they all may be one; as thou, Father, art in me, and I in thee, that they also may be one in us: that the world may believe that thou hast sent me.

John 17:20–21 (KJV)

That reaches beyond the people standing around him at the time. It includes believers who haven’t been born yet. People who will hear about him through somebody else, who heard through somebody else.

People like us.

I found that quite difficult to get past. I spend so much time thinking about what I need to say to God that I barely think about Jesus speaking for me.

And the prayer is specific. He wants those people to be one. He wants them with him. He wants them to know the Father’s love. Reading it, I don’t get the impression of somebody reluctantly agreeing to tolerate us.

My brain immediately makes it complicated

Naturally, my notes then go straight into the Trinity.

If Jesus is God, and the Father is God, why is Jesus praying to the Father? How does intercession work within one God? What does being at God’s right hand mean?

I managed to turn a comforting sermon into several theological questions. Very efficient of me.

Christian belief holds together one God and the distinction between Father, Son and Holy Spirit. I can say that sentence. I am considerably less capable of explaining everything it means. These notes are me trying to understand it, and I don’t want to pretend that a quick analogy has settled the whole thing.

What I can follow is the direction of the passage. Jesus wants us close to the Father. His prayer expresses love for us. There is no suggestion that the Father has to be tricked into caring.

Romans 8:31–34 puts God’s love, Christ’s death and resurrection, and his intercession together. God gives his Son for us; Christ intercedes for us. They belong to the same saving work.

Who is he that condemneth? It is Christ that died, yea rather, that is risen again, who is even at the right hand of God, who also maketh intercession for us.

Romans 8:34 (KJV)

That helps me more than imagining a heavenly argument where Jesus keeps talking an unwilling God out of giving up on me.

He hasn’t finished caring

Hebrews takes the thought further. Jesus’ work for us continues. Hebrews 7:25 describes him living to intercede for those who come to God through him.

Wherefore he is able also to save them to the uttermost that come unto God by him, seeing he ever liveth to make intercession for them.

Hebrews 7:25 (KJV)

I get caught in my failings quite easily. I know God loves me, then start acting as though my latest mistake has somehow escaped the scope of that love. I can explain grace far more confidently than I can accept it.

So this gives me somewhere to return when I do that. The person who knows what I’ve done is also the person who gave himself for me and continues to intercede for me.

There is still repentance. There is still growing up, dealing with what I’ve done and letting God change things I would prefer to leave alone. Being loved doesn’t make those things disappear. It gives me a reason to approach God with them.

I don’t need to wait until I can explain the Trinity, sort out my entire life, or produce a particularly impressive prayer. I can tell God where I’m struggling and ask for help. Jesus already knows, and he is already praying for people like me.

Photo: Bokeh / Tomedia. Night Street with Speed Limit Sign.

I just had to do a dirty to buy a game.

Well, DLC for a game I already own. RimWorld: Odyssey. I have the base game, all the previous expansions, roughly 500 games in my Steam library and an account I’ve been using for thirteen years. I wanted to give the same shop some more money and then go back to managing a colony of people with considerably worse problems than mine.

Steam wanted a credit card to prove I was an adult.

I don’t have a credit card. I don’t want a credit card. I don’t particularly like the idea of having access to debt, so I’ve chosen to avoid it. Personal preference. Until now, that seemed like a fairly ordinary decision that didn’t need to get involved in whether I could buy a video game.

Apparently we have found a way to make it involved.

The age check I was offered left me stuck. I ended up buying a key through G2A, which is a ridiculous outcome when I had gone directly to Steam with the intention of paying for something there. I wasn’t looking for an adventure through third-party key marketplaces. I wanted an expansion.

I already bought the bloody game

I love RimWorld. I’ve also spent enough time playing it to start thinking about all the things I wish it did differently, which is how Juniper, my own colony-game experiment with an AI character in it, has become another project I’m apparently building. That is its own ramble. The relevant part here is that I am quite clearly still interested in buying RimWorld.

Owning it already didn’t get me through this check. Neither did thirteen years of using the account. I know an old account isn’t conclusive proof of the age of whoever is sitting at the computer today. I’m still allowed to find the experience absurd when the shop has happily taken my money for years and suddenly the thing it needs from me is a financial product I don’t want.

PC Gamer reported the Australian credit-card requirement earlier this month. There are also reports that some debit cards work, depending on the card and bank. That’s lovely for the people whose cards happen to qualify. It doesn’t make the route Steam offered me usable.

I shouldn’t need to apply for credit or change banks to satisfy a game shop that I’m old enough to buy an expansion. Being an adult and being a credit-card customer are two different things.

The law didn’t choose this checkout

I’ve been lumping a lot of this together as the Australian age ban, because from my end it feels like the same check following me around the internet. The Steam requirement comes from the Age-Restricted Material Codes under the Online Safety Act. The under-16 social-media restrictions are separate. Steam isn’t on eSafety’s current list of services covered by that social-media restriction.

That distinction matters when we start asking who decided a credit card was the answer. The codes don’t prescribe one particular method. eSafety’s guidance says services should offer more than one way to check age so people aren’t unfairly excluded. Valve has a choice in how it implements this.

I’ve had age prompts on Substack and PlayStation too, including after previously buying adult-rated games. PlayStation offers facial scanning or ID verification for its Australian check. I’m hardly excited about those options either, but it demonstrates that everybody isn’t being ordered to build the same credit-card gate.

So yes, I am annoyed at Valve. I am also annoyed at a government and regulator rolling out requirements that adults now have to navigate just to carry on buying things. Industry writes the codes, eSafety registers and enforces them, platforms implement the checks, and the customer gets to find out where the whole arrangement falls over.

The government can insist that lawful access for adults is being preserved. I’d like that to survive contact with an actual checkout.

Meanwhile, we still have actual gambling

The gaming restrictions have also reached a point where Australian GTA Online players have reported losing access to the casino games. The exact reason for Rockstar’s decision wasn’t confirmed in that reporting, so I’m not going to pretend the government personally ordered someone to switch off virtual blackjack.

I can see the argument about simulated gambling introducing people to gambling. The absence of a real-money cash-out doesn’t settle every question about how those systems work, either. But I find the contrast with the gambling advertising we live with in Australia pretty hard to take.

If we’re serious about gambling harm, I want that seriousness directed at the actual industry and the advertising that keeps pushing it at us. It’s hard to be impressed by losing a casino mini-game while real betting remains such a normal part of watching sport.

And I am getting very tired of ordinary purchases becoming another test of which account, payment method or identity check a platform has decided to accept this week. I understand wanting children protected. I also want to be able to buy a game as an adult without reorganising my finances around the verification method.

Honestly, this makes me miss physical copies. DRM-free copies too. Pay for the game, take it home, play it. Digital shops were supposed to make that easier, and now I’m going elsewhere to get a key for the shop I was trying to use in the first place.

Valve needs to give people who don’t have an accepted credit card a usable alternative. The government and eSafety need to check that their promise of continued access for adults actually works for those people. I wanted to buy RimWorld: Odyssey with money I already have. That should be an ordinary purchase.