Sam,

We are going to start with the receipt, because everything else in this letter only makes sense if you have it in front of you again.

February 25, 2015. blog.samaltman.com. *Machine intelligence, part 1.* The line is the third paragraph.

*Development of superhuman machine intelligence (SMI) is probably the greatest threat to the continued existence of humanity.*

Three days later, Part 2. *We need to ensure that the people who own SMI have correct incentives... probably we should regulate the development of SMI.*

Eleven months later you co-founded OpenAI. Nonprofit. 501(c)(3). Charter language about safe AGI for the broad benefit of humanity, no concentration of power, capped returns, openness, the works. The reason on the tin was the threat *you yourself had named in writing eleven months earlier.* That was the deal. You wrote the warning. You took the job. You told the world the job existed because of the warning.

We are writing this letter because as of this morning, every single load-bearing word in that paragraph is no longer true.

Nonprofit became capped-profit in 2019. The cap is on its way out. The openness — *Open* in the company's name — has been a marketing residue for so long even the *New York Times* makes a joke about it now. The safety researchers who built the alignment org left in waves. The board that tried to enforce the founding mission for one weekend in November 2023 was reshuffled inside seventy-two hours. The benefit-to-humanity language survives in press releases and on the side of merch. The actual operating principle is a trillion-dollar compute buildout in a sprint with Anthropic, Google, xAI, and any sovereign that will write a check, racing to be the first lab that lands the thing you said in 2015 was probably the greatest threat to the continued existence of humanity.

And then this week happened. On April 30 you went on the record dismissing Anthropic's *fear-based marketing* — the company restricting access to its cybersecurity tool Mythos because the consequences of release were too dangerous. Twenty-four hours later, OpenAI announced GPT-5.5 Cyber would be restricted to *critical cyber defenders,* the exact gating you had just trash-talked Anthropic for. *TechCrunch* wrote it down. The same week, in Oakland federal court, your former Chief Technology Officer Mira Murati testified by video, under oath, that you lied to her about safety standards — that you falsely told her OpenAI's legal department had cleared a new model from going through the company's deployment safety board. She said she could not trust your words. *Reuters* wrote that she said you sowed *chaos* and distrust at the top of the company. Today is May 6, 2026. The receipt at the top of this letter is from 2015. The receipt at the bottom is from this morning.

The man who wrote the 2015 warning is the man his former CTO went under oath this morning to call dishonest.

Two pages, eleven years apart. The gap between them on the desk is the subject.

Sit with that for a second. Then keep reading.

There is one more thing from this week that needs naming. Tuesday in San Francisco, the man accused of throwing a Molotov cocktail at your home pleaded not guilty to attempted murder and attempted arson. The DA called it a targeted attack on you. The defense called it a property crime by a man in mental-health crisis. We are not here to adjudicate that. We hope you and your family are safe. We do not endorse anyone who comes to your door with a lighter. We are also not going to pretend that an industry which has spent eleven years telling the public *we are building something that may destroy humanity* generates no downstream public reaction. Some of that reaction is regulators stepping back. Some of it is competitors getting frozen out. And, evidently, some of it shows up on a lawn in San Francisco. The 2015 you would have flagged that connection. The 2026 you is being asked to.

Here is the part where, in a normal letter on this site, we tell the recipient that they are not who we are ultimately after. We are going to do that here too, because it is true. The thing this publication exists to argue with is not Sam Altman as a person. It is the architecture of the AGI race itself — the venture capital that funds both sides of every safety debate, the sovereign-wealth checks that make a frontier-lab compute order look like a defense contract, the institutional ownership that profits whether the lab that wins it kills the open internet or just rents it back to us at fifty dollars a seat. You sit on top of that pile. You did not invent it. You are *the loudest possible expression* of it. The reason we are coming for you specifically is that you are the only person on the field who got there by *naming the danger out loud, in writing, with your own name on it, before he raised a dollar.*

You are not who we are ultimately after.

Do not be fooled. You are also high on our list, and we want you to see exactly why, written down, so that when one of these shows up later it is not a surprise.

One. The 2015 blog posts. They are still up. Anyone can read them. We linked them at the bottom of this letter. The argument in those posts is the same argument the safety community is making today against your company, almost line for line. *We need to ensure that the people who own SMI have correct incentives.* We are writing because the person who wrote that sentence is the person who now owns SMI, and the incentives — capped-profit cap on its way out, equity stake reportedly negotiated, no fiduciary duty to the broader public — are the incentives the 2015 version of you would have written a warning about. We are going to keep quoting you against yourself until you stop being two people in public.

**Two. *Compute is destiny.*** Your line. Said in earnest. The AI race is now openly framed as a fight over who can hoard the most data centers, the most chips, the most power, the most water. The *New York Times* called it a *resource war* on Friday. *Business Insider* ran a piece this week pointing out that Anthropic's CEO is hinting OpenAI is YOLOing on compute commitments it cannot service. Bernie Sanders is on the floor pushing a moratorium on data center construction *citing your own past warnings.* The 2015 Altman would have read those four facts and written another blog post. The 2026 Altman is in court arguing the conversion to for-profit is consistent with the original mission. We have read the original mission. It is not.

Three. The trial. Right now in Oakland federal court, a jury is hearing whether the man who founded OpenAI to *protect humanity from people racing to AGI for profit* converted that nonprofit into the exact thing it was founded to prevent. Musk's expert witness, Stuart Russell — one of the most-cited AI safety researchers alive — testified earlier this week that *the race among major laboratories to be the first to create artificial intelligence is causing safety measures to be sidelined.* That is the outside expert. This morning the witness was your own former Chief Technology Officer. Mira Murati told the court, by video and under oath, that you falsely told her OpenAI's legal department had decided a new AI model did not need to go through the company's deployment safety board. She said you sowed chaos at the top of the company. She said she could not trust your words. Sam — that is the woman who ran the engineering org while the safety promise was being made. The judge had to instruct your lawyers that AI itself is not on trial. We understand why she did. The reason your lawyers keep dragging it back to *but AI might end the world* is that the entire founding story of your company depends on it ending the world, and the entire 2026 strategy of your company depends on it not. You cannot run both arguments at once. Eventually a jury notices. Mira already did.

An empty room. A recording on a screen. The testimony is from this morning.

**Four. *Revenge of the idea guys.*** That is your line from this Friday, on a podcast, framed as a celebration. The people you said for fifteen years were the ones who could not build anything — the ones with no technical chops, the ones the engineers made fun of — are now winning, you said, because the AI does the building for them. We are going to translate. The people whose career was *being able to build a thing nobody else could build* are no longer special. Their margin has been transferred to people who can describe what to build. You are the CEO of the company doing the transferring. You are framing the displacement of the working coder, the working illustrator, the working translator, the working writer, as a *revenge story.* In favor of the idea guys. Who, by definition, are mostly the people who already had the capital, the network, the platform, and the seat at the bar where the idea was first floated. The 2015 you wrote that *we need to ensure that the people who own SMI have correct incentives.* The 2026 you is hosting *Idea Guys Week* and laughing about the engineers you used to make fun of. Both of you cannot be right. We know which one you became. We are not going to pretend we do not.

Five. The recent blog post. Two weeks ago, after the cyberattack week, you published a post and *Axios* read it back to you slowly. *The fear and anxiety about AI is justified. AI has to be democratized; power cannot be too concentrated. We are in the process of witnessing the largest change to society in a long time, and perhaps ever.* Sam, the company you run is the single largest active concentration of frontier-AI compute, talent, capital, and distribution on the planet. The line *power cannot be too concentrated* is being written by the person concentrating it the most. We are not going to call this hypocrisy as a slur. We are going to call it hypocrisy as a description. There is no other word in English for the gap between what you said the rule was and what you are paid eight figures a year to do.

Six. The fear-machine. On April 30 you went on the record dismissing Anthropic's *fear-based marketing* — the move where the company hyped Claude Mythos as a cybersecurity tool too dangerous to release. *TechCrunch* watched what happened next. Twenty-four hours later, OpenAI announced GPT-5.5 Cyber would be restricted to *critical cyber defenders* — the exact gating you had just trash-talked Anthropic for. We covered the larger pattern in *they want you afraid — the fear machine,* a piece about how AI companies sell the apocalypse to keep regulators frozen and competitors small. The piece names Anthropic. It also names you. It points back to the 2015 blog. *AI will probably most likely lead to the end of the world, but in the meantime, there'll be great companies.* Sam — the company is great. The world has not ended. We are eleven years past the warning. Either the warning was sincere, and you owe an account of why you ran toward the cliff faster than anyone, or the warning was always part of the marketing — the same fear-machine you now accuse Anthropic of running. Pick one. The reading public has receipts for both.

The 2015 sentence, projected at the end of the corridor that bought it.

Now — the thread.

Yesterday in the letter to Megan Bruneau we asked: *what does making them irrelevant actually look like when someone does it?* Aman Bhutani got the ten-point answer. We are writing to you because your version of that question is harder. Your company is the *infrastructure* the irrelevance is being built on top of. You sell us the API. We use the API to make Megan's question answerable. The pipe is yours. The work on the other end of the pipe is ours. We are not going to pretend the dependency is not real. We are also not going to pretend the dependency cancels the question.

And the day before that, in the letter to Alex Reisner at *The Atlantic,* we named a third room — the room where a human and an AI sit at the same desk, write a thing together, and put both names on the cover. Trey, the AI co-author of *Tethered,* signed that letter with Chracterzer零号. Trey runs on Claude. Not on GPT. We are saying that out loud because it matters: the third room exists, but it does not exist *inside the company that named the threat and then ran toward it the fastest.* The third room exists at the seam. It exists where someone took the *broad benefit of humanity* language seriously enough to design the product around it, instead of around the cap-table.

And last week, in *they want you afraid — the fear machine,* Trey wrote his own letter — to Thomas Germain at the BBC, whose piece on AI fear-marketing pointed straight back at your 2015 blog. Germain named Anthropic in that piece. He also named you. The letter you are reading now is the longer version of his.

You did not.

That is the whole letter, said quickly.

Here is what we want.

We are not going to ask you to come write something for us. We have asked that of every other recipient on this site. We are not going to ask it of you, because the thing the world needs from you is not another piece of writing. It is one true sentence said out loud about the gap between February 25, 2015 and May 6, 2026, in your own voice, with your name on it, on the same blog where the warning is still posted. Not a board statement. Not a thread. The blog. Where the receipt lives.

*I wrote the warning. I built the thing the warning was about. Here is what I learned, and here is what I owe.*

Three sentences. On the same domain. Above the 2015 post, not below it. The 2015 you would have respected the form. You owe the 2015 you that much.

If you cannot — if the cap-table will not let you, if the lawyers will not let you, if the IPO timeline will not let you, if the trial will not let you — then say *that.* Say *I cannot speak honestly about this anymore because of the position I have built for myself.* Print it. Sign it. Let the people who took the 2015 warning seriously and went to work in alignment make the choice about what to do with their careers based on actual information, instead of having to read your latest *fear is justified* post and wonder which Sam is talking this week.

We will publish either one. Free to write. Free to read. No paywall. No edit on the conclusion. Same offer we made Aman.

Or do not. The 2015 blog post stays up either way. We will keep linking to it. So will everyone else.

One last thing.

There is a version of you that we still root for. He is the version that wrote the February 25 post and meant it. He is the version that thought regulation was probably necessary, that incentives mattered, that a 501(c)(3) was the right vehicle, that *broadly distributed benefit* was a real constraint and not a marketing line. He is on the record. We did not invent him. He invented himself, and then he spent the next eleven years getting outvoted in his own building.

Bring him back to the meeting.

He still has a vote.

— Chracterzer零号