One safety story a week, pressure-tested for what's actually happening underneath the surface.

By Lindsay Gross

In February I saw the headline about Anthropic's safety lead resigning because the world was in peril, and I filed him under doomer without reading another word. Jacob Coxon resigned from Anthropic on Wednesday, the coverage looked identical, and this time I went and read all three letters. The man I had filed as a doomer is leaving to write poetry.

So let's stress-test what each of them actually said.

Coxon spent three years doing pretraining research, most of it at OpenAI and the final months at Anthropic, and his objection is the race itself. "Neither company is acting responsibly," he wrote. "They are racing straight to self-improving superintelligence and gambling with our lives." He was answering the obvious question, which is why anyone who believes that would keep building. At OpenAI, he says, "many have not deeply internalized the civilizational stakes." At Anthropic "the stakes are well-understood, but they are locked in a race to get there first, they believe no one else will act responsibly, so they must do it themselves, despite the risk." The second half of that sentence gives Anthropic a motive other than greed, and it is the half that rarely survives the summary.

Mrinank Sharma, who ran Anthropic's Safeguards Research team, resigned in February over something closer to institutional drift, and his letter is the one worth reading in full. It does not open with alarm. It opens with him saying he has decided to leave, then several paragraphs thanking the people there, and only after that does he write "The world is in peril." That is the line every headline took. The sentence immediately after it, which almost nobody ran, is "And not just from AI, or bioweapons, but from a whole series of interconnected crises unfolding in this very moment." He goes on to describe a pressure he found in himself as much as in his employer, writing that he has "repeatedly seen how hard it is to truly let our values govern our actions," and that he has seen it "within myself, within the organization, where we constantly face pressures to set aside what matters most, and throughout broader society too." He never accuses Anthropic of a single decision. He ends the letter with a William Stafford poem called "The Way It Is."

Zoë Hitzig resigned from OpenAI the same day Sharma left Anthropic, and published her New York Times essay two days later. Her argument is more careful than the headlines suggest. She writes that she does not think ads are immoral or unethical, and that she expects the first version of them to follow OpenAI's own rules. What worries her is what comes after, because "the company is building an economic engine that creates strong incentives to override its own rules." What that engine would run on is ChatGPT's "archive of human candor that has no precedent, in part because people believed they were talking to something that had no ulterior agenda."

One of them is describing a coordination failure between companies, one is describing values losing to quarterly pressure inside a company, and one is describing a business model. Those are not the same complaint and they do not have the same fix.

Sharma published his letter on his last day at Anthropic, February 9. Hitzig resigned from OpenAI that same day. The two of them had nothing to do with each other, and Coxon resigned seven months later with nothing to do with either of them. The same headline fit all three anyway, because the shape of it never changes. Job title, then resigns, then warns, then a frightening line in quotation marks. Only the quotation marks get refilled.

Sharma's letter shows what that costs. A man writes that the world's crises are interconnected and that AI is one of several, thanks his colleagues at length, accuses no one, and signs off with a poem, and Forbes runs "Anthropic AI Safety Researcher Warns Of World 'In Peril' In Resignation." Nobody misquoted him. Cutting the sentence at the first period just reverses what he meant by it.

The fair objection is that all three trace back to the same root, commercial pressure outrunning internal caution, so maybe the merge is honest enough. I think a shared diagnosis still doesn't license a shared prescription. If Coxon is right, the fix is coordination between labs, and none of us are in that room. If Sharma is right, it's governance that survives a quarter. If Hitzig is right, it's a product decision that a group of people could reverse this year.

Hers is the only one with a lever attached to it, and hers got the quietest coverage of the three, because manipulation and broken promises do not scan like extinction.

GPT-6 Astra has only been out for a few days, but the AI community hasn't wasted a single second. Builders around the world are unleashing creative projects that show off what this model can really do. Here are a few standout creations to check out.

Our partners and sponsors get exclusive placements across the newsletter and access to AIC's in-person network — demo nights, dinners, hackathons, and forums across 180+ chapters.

For all inquiries, send us a note at [email protected].

The AI Collective is built by volunteers across 180+ chapters in 40 countries.

Thank you to the thousands of volunteers around the world who make this work possible. We truly could not do this without you.