TLDR
Jacob Coxon resigned from Anthropic on 9 September 2026, two months before his shares vested, accusing both Anthropic and OpenAI of racing toward self-improving AI without adequate safeguards. OpenAI announced alignment veteran Paul Christiano's appointment to its Safety and Security Committee on the same day.
KEY TAKEAWAYS
What Coxon said, and what he left out
Jacob Coxon posted his resignation thread on X at the start of a Wednesday morning and did not waste words. "They are racing straight to self-improving superintelligence and gambling with our lives," he wrote, after three years doing pretraining research across OpenAI and Anthropic.[1] The sentence was built to travel, and it did.
Read carefully, the thread does not identify a single concrete incident that pushed Coxon out the door.[1] There is no leaked memo, no refused safety review, no model deployed against his recommendation. Coxon describes a direction of travel, not a specific event, and that gap matters for anyone trying to weigh the warning.
Coxon did put forward concrete proposals. He called for pacing agreements between US labs and floated a temporary ban on capability improvements, citing "warning shots like the Hugging Face attack" as evidence that coordination between competitors is at least conceivable.[1] Those are specific policy positions, which puts him ahead of most public AI critics in terms of offering something actionable.
I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.
2026-09-09 · View on XWhy the equity detail matters more than the headline
Coxon quit two months before his equity would have vested, forfeiting those unvested shares entirely.[2] Public warnings from AI insiders are common enough to have their own genre; most cost the person who makes them very little beyond some social awkwardness.
Forfeiting equity is different. It is a verifiable financial commitment, hard to fake and hard to walk back. It does not prove his specific claims about either lab's safety posture, but it does confirm that Coxon believed what he wrote enough to pay for saying it, which moves him out of the category of people performing concern.
How this compares with the 2024 OpenAI departures
The pattern Coxon's resignation fits is not new. In 2024, a sequence of high-profile exits from OpenAI, including Ilya Sutskever and members of the superalignment team, generated a comparable cycle: insider warning, headlines, governance announcement, capability development continues. OpenAI's response at the time was to establish an independent Safety and Security Committee to oversee model development and deployment, publishing that update in September 2024.[3]
Anthropic and OpenAI both launched with charters emphasising safety and public benefit, and both have since released progressively more capable models under commercial and geopolitical pressure. The tension Coxon describes is structural, built into the founding conditions of both organisations, which makes his warning credible as a description of the environment while leaving open the question of whether anything specific has changed to make now more dangerous than six months ago.
The Christiano appointment: genuine counter-move or optics?
On the same day Coxon resigned, OpenAI announced that alignment veteran Paul Christiano would join the OpenAI Foundation Board and its Safety and Security Committee as a non-voting observer.[4] The timing allows two equally reasonable readings, and the facts alone do not settle which is correct.
Bret Taylor, Chair of the OpenAI Foundation and OpenAI Group PBC Boards, said Christiano has helped define the field of AI alignment through work that is rigorous and focused on the hardest questions posed by increasingly capable systems, and that his experience across frontier safety and government standards work will strengthen the Board's oversight and the Safety and Security Committee's work.[4]
Christiano's own statement was less settled than Taylor's framing. Christiano said AI capabilities have advanced very rapidly in the last year and alignment remains a difficult technical problem, making the Safety and Security Committee's responsibility more important and more challenging than ever, and that he is excited to join the team to help.[4] That reads more like an acknowledgement that the problem is getting harder faster than the solutions are arriving.
OpenAI's October 2025 recapitalisation restructured its governance to give the nonprofit Foundation formal control over safety oversight via that same committee, a change positioned at the time as exactly the kind of structural safeguard critics had been asking for. Whether the Christiano appointment represents that committee functioning as designed, or represents OpenAI managing the optics of a difficult news day, the two possibilities sit directly on top of each other. What is verifiable is that Coxon's resignation and Christiano's appointment share a date, and labs in this position have a consistent record of responding to insider departures with governance announcements rather than capability slowdowns.
SOURCES & CITATIONS
FREQUENTLY ASKED QUESTIONS
Why did Jacob Coxon resign from Anthropic?
What did Coxon forfeit by resigning when he did?
What is Paul Christiano's role at OpenAI?
What specific policy changes did Coxon propose?
Xaviery Malinao writes for Prompt the Market on how brands and agencies are adapting to answer engines, drawing on Bushnote's work with clients across search, AI search and content.
