Jacob Coxon, a researcher at Anthropic who previously worked at OpenAI, resigned and posted that neither company is acting responsibly, that the leading labs are sprinting to build systems capable of hacking anything and acquiring real power and resources without appropriate safeguards, and that if the pace does not slow there is a strong chance we could all die in the immediate future. The post reportedly cleared eighteen million views. He was speaking with the BBC within days, describing colleagues who are genuinely frightened, who are planning what to do with their lives, some of whom are considering buying land somewhere because they are afraid of the instability ahead. Evan Hubinger, who works on alignment at the same company, wrote publicly that he puts the odds of this technology eliminating humanity above ten percent within the decade.

Days after that, Dario Amodei, the chief executive of Anthropic, published a thirty-eight-hundred-word essay calling for a global slowdown in capability advancement, third-party assessments from embedded evaluators, coordination among democratic governments, and eventually coordination with authoritarian ones. Sam Altman of OpenAI agreed within hours. Elon Musk posted three words of agreement. Demis Hassabis of Google DeepMind called it a step in the right direction. An Anthropic spokeswoman said the company has always been transparent about both the benefits and the risks and continues to build models with some of the strongest safeguards in the industry.

I am going to record one more piece of context and then move on from it, because it is not the center of what I want to say.

Anthropic is reportedly preparing for a potentially record-setting public offering. Jensen Huang, the chief executive of Nvidia, said at a San Francisco conference that labs were escalating security concerns because they intended to sell security products, and asked what better way there is to create demand than to create a problem. Clément Delangue, the chief executive of Hugging Face, said it is imperative that safety not be settled behind the closed doors of a handful of frontier labs. Those are not my characterizations. They are on the record from people inside the industry.

What the legislature did with a warning about human extinction.

Speaker Mike Johnson, Republican of Louisiana, said the House would not rush into legislation that could put China at an edge, and that the first step is a meeting of all the platform providers. Representative Hakeem Jeffries, Democrat of New York, criticized the pace, called for decisive action, and when asked which specific restrictions he would support, said the conversation would happen that week. Representative Anna Paulina Luna, Republican of Florida, called for a special session. Representative Lori Trahan, Democrat of Massachusetts, said safety researchers are resigning and companies are racing ahead anyway and it is past time for Congress to do its job. Senator Ruben Gallego, Democrat of Arizona, dismissed the China framing outright, pointing out that a genuinely dangerous system will not care about anyone's nationality. Kevin Hassett, director of the White House National Economic Council, said the administration was studying a proposed kill switch. President Trump told reporters he has no concerns beyond falling behind.

Former President Barack Obama, at a closed-door fundraiser in Manhattan, urged Democrats to build a framework and told them he considers himself neither an accelerationist nor a doomer.

So: the men who built the thing announced that the thing might kill everyone, and the institutional response from the legislative branch of the United States was a proposal to convene a meeting with the men who built the thing.

That is not a regulatory posture. That is a hearing invitation extended to the only witnesses anyone thinks to call.

Dr. Timnit Gebru has been saying this. She has been saying it for years.

Not the extinction part. The part underneath it, which is the part that is actually load-bearing.

She wrote this week that it feels like being hit by hype after hype in the run-up to the OpenAI and Anthropic offerings, one thing after another while the companies try to one-up each other. She described finishing an op-ed about the week's headlines and then immediately seeing the viral resignation post land on top of it. She said she has been making the same argument technically, historically, philosophically, and by analogy, for a long while now.

And then she said the thing I want carried furthest. She asked why these narratives intensify precisely now, and answered her own question. If you are frightened by how powerful the product is, the fear markets the product and sidesteps the accountability in a single motion, because then the problem is no longer corporate negligence. The problem is the model going rogue. Not the theft, not the labor exploitation, not the environmental cost, not the incompetence. The machine.

Her account of what is genuinely existential is concrete and checkable. Autonomous weapons already being used in warfare. A climate catastrophe these buildouts accelerate. Employers using this technology as cover to shed workers. She has called the machine god narrative a distraction from all of it.

You do not have to agree with every line of that to notice that it is a serious argument, offered for years, by someone with the credentials to make it, and that it has never once produced the week of coverage that a single resignation post produced.

The credential is not expertise. It is proximity.

I made a video recently about what I have been calling the credentialing of alarm, which is the observation that the argument does not have to improve in order to be heard. The speaker has to get credentialed, and the credential that qualifies a warning in this field is proximity to the capital that built the thing being warned about.

This week was that pattern running at full speed, and it was worse than the version I described before, because the comparison is now direct.

Dr. Timnit Gebru and Dr. Margaret Mitchell documented the risks of scale at Google and lost their positions over it. Dr. Joy Buolamwini documented how these systems fail on Black faces, and a major corporation mounted a public campaign to discredit her research. Bill Gates, the co-founder of Microsoft, gave an hour to The New York Times a few weeks ago saying a substantial portion of this out loud, and it did not move anything resembling eighteen million views.

One young white man resigned, and the Speaker of the House started scheduling.

I want to be careful and exact here, because precision is the entire discipline and I am not interested in attacking a person I have never met. Jacob Coxon appears sincere. He described the decision as a gradual emotional accumulation that eventually snapped. I have no reason to doubt him and no interest in impugning him.

What I am describing is not about him. It is about the machinery that determined the reception. The identical sentence carries radically different weight depending on who holds it, and in this country the weight accrues to youth, to whiteness, and to proximity to the people writing checks. Black women said it first, said it with citations, and paid for saying it. The field absorbed their argument and discarded the authors, and is now hearing it back, five years later, from someone it finds easier to believe.

That is not a coincidence of timing. That is a selection system, operating exactly as it was built to operate.

And the framing itself is doing work.

There is a clarifying argument from Professor Virginia Dignum that I want in this piece, because it cuts the knot.

She has written that headlines about systems taking control are irresponsible, because they suggest a will that does not exist and shift attention away from the people who are actually responsible toward something that cannot be held to account. Hacking, acquiring resources, and self-improvement are actions in the world. They require compute, energy, money, and access rights, and every one of those is granted by people. Deploying agents with broad permissions and no oversight, she argues, is a governance failure rather than an emergent property of the technology. Agents do not escape. They are given a goal without constraints, and they pursue it.

She also makes the point about the percentages that nobody making them wants to sit with. A probability estimate requires a model of the process, a reference class, and data to calibrate against. For this event, none of those exist. What gets reported as ten or twenty or ninety-nine percent is a personal belief, usually elicited from people with a professional or commercial stake in the answer, and it cannot be checked against anything.

Meanwhile the documented harms are sitting right there, unglamorous and enormous. The Dutch childcare benefits scandal, where an automated risk system wrongly accused tens of thousands of families, disproportionately families with a migration background, of fraud, producing debt and lost homes and children taken into care, and bringing down a government. Australia's Robodebt scheme, which issued hundreds of thousands of unlawful debt notices through automated income averaging and was found by a Royal Commission to have caused serious harm including deaths. The United Kingdom Post Office, which prosecuted hundreds of sub-postmasters on the output of faulty accounting software that nobody was willing to question.

In every one of those, the software was software. The decisions were made by people who deployed it without oversight or recourse. The victims were real, they were identifiable, and they were not hypothetical.

The extinction narrative pushes all of it off the front page. That is not an accident of news cycles. It is the function.

So where does this leave us.

It leaves us with a governance question that nobody in that entire week of coverage bothered to ask.

Every permission these systems hold was granted by a person inside an organization. Every one of those grants can be bounded, logged, audited, revoked. When a technology is described as an autonomous threat rather than a sequence of choices, accountability evaporates, because a danger that originates in the machine is a danger no board of directors has to answer for. That is the structural benefit of the story, and it accrues to the storyteller.

Which is exactly why it matters who is in the room.

Not consulted. Not invited to comment during a window. Seated, with reserved standing. With their documented experience admitted as evidence rather than heard and set aside as anecdote. And with somebody accountable to them holding the actual power to stop a deployment that is not working, rather than the power to write a strongly worded assessment after the fact.

The people already living inside the harm have been describing it accurately for a decade. What has never been built is the infrastructure that would make their description count for anything. And every week that looks like this one is another week it does not get built, because the conversation is taking place between the companies and the Congress, and the Speaker's proposed first step is to invite the companies.

Dr. Ruha Benjamin, in Race After Technology, described rusty value judgments embedded in shiny new systems.

Eight words, and they hold the entire week.

The technology is new. The question of whose alarm counts, whose testimony is evidence, and who gets to be the person the country finally listens to is not new at all. It is the oldest question we have, wearing this season's clothes, and we are about to answer it the same way we have answered it every other time unless somebody builds the thing that makes a different answer possible.