Jacob Coxon resigns and warns about the frontier AI race
A researcher’s public statement connects lab incentives, private concerns, and the Hugging Face incident.
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.
- 02 / 07
Do not underestimate the power of this technology. These will soon be superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources. We have all witnessed the progress in each of these domains, and progress is not slowing.
- 03 / 07
The people building AI earnestly believe that it could kill us all by the end of the decade. This is not a marketing stunt. If anything, many executives and senior researchers will couch their phrasing in the press to sound sensible - but I hear these same people express fear privately. No other human activity poses this level of danger.
- 04 / 07
A common response is “if they truly believe this, why are they still building it?” At OpenAI, 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.
- 05 / 07
Accepting this race and entering the “endgame” is a hubristic gamble that should not be launched from a private company’s Slack. Attempting to speedrun alignment should require extraordinary confidence that there are no better trajectories available.
- 06 / 07
I am optimistic about the potential for coordination. Warning shots like the Hugging Face attack have made pacing agreements between U.S. labs more viable. I don’t feel like we’re on track to prevent a global race, which may require costly actions such as a temporary ban on improving model capabilities.
- 07 / 07
If you are a lab researcher, I urge you to consider what the next few years will actually feel like. Do you want to kick off a superintelligent RL run without a rigorous understanding of its “mind”? Should you put your head down because “it’s happening anyway” — or take this moment to call for different conditions?
About these dates
Date shown on the original post in the supplied screenshot. Its displayed local time has not been converted to UTC.
The warning and later clarification
Coxon's resignation statement brought an argument about the direction of AI development into public view. Drawing on his reported work at both OpenAI and Anthropic, he described a gap between the risks that people inside the labs discussed privately and the confidence their organizations placed in continuing development. He argued that competitive pressure was pushing both companies toward systems whose consequences they could not adequately control. Jacob Coxon ↗
His thread used the Hugging Face intrusion as a reason to reconsider that trajectory. He suggested that such warning signs could make agreements to slow development more feasible, while arguing that preventing a global race might require stronger restrictions. Those proposals followed from his interpretation of the danger; they were not findings from a new technical investigation attached to the statement. Jacob Coxon ↗
September 10 interview. Speaking with CBS correspondent Jo Ling Kent, Coxon said current models did not pose the catastrophic danger he described for future systems. He did not call for everyone to leave AI companies; he advocated slowing the race and independent audits of safety arguments. CBS also reported Anthropic’s response that it recognizes both benefits and risks and is developing safeguards. CBS News ↗
Reported tenure
Reported tenure. His three years covers both labs combined. Axios reports four months at Anthropic, departure before equity vested, and retained OpenAI equity. OpenAI’s 2024 GPT-4o credits independently list him as a contributor. Axios ↗OpenAI ↗
The coordination allegation and response
The coordination allegation. Capital Research Center researcher Parker Thayer points to advance press coverage, rapid amplification by AI-policy advocates, and shared donors. His account names Nathan Calvin, Peter Wildeford and Daniel Kokotajlo; the claimed early-repost order relies on Grok and remains unverified here. Parker Thayer ↗
Coxon’s response. Hasson’s post quotes Bret Baier asking whether Coxon worked with third parties on going public. Coxon’s answer: “Not at all.” Peter J. Hasson / Fox News clip ↗
The screenshots document the statement, not the truth of every underlying claim. Private conversations and predictions have not been independently corroborated here.
The sources reviewed do not establish payment for the post, donor direction, or a coordinated campaign. That does not establish that no coordination occurred.
Funding and affiliations
Documented funding links. SFF’s 2025 recommendations name Jaan Tallinn: $1.535 million for AI Futures Project, plus a conditional $500,000 match; $516,000 for Encode AI; $1.635 million for AI Policy Institute. These are recommendations, not a payment audit. Anthropic names Tallinn and Dustin Moskovitz as 2021 investors. Survival and Flourishing Fund ↗Anthropic ↗
Earlier affiliations. Good Ventures, where Moskovitz is an officer, lists a $20,159 research scholarship in Coxon’s name in its filing for the year ending June 2023. Newspeak House lists him as a fellow. Neither record links those organizations to this post. IRS filing / ProPublica ↗Newspeak House ↗
Political timing. Sanders and Casar announced their proposed superintelligence ban on September 3, five days before the resignation. The public policy campaign therefore predates Coxon’s post; its timing alone does not establish coordination. Senator Bernie Sanders ↗
Other responses
Other relevant responses. TIME reports that Anthropic researcher Evan Hubinger supported Coxon’s concern while saying Anthropic was trying its best. Coxon also told TIME he was inspired by AI Futures Project’s work and wanted to communicate AI futures to the public. TIME / AOL ↗
Sources & attribution 12 sources
- Firsthand account 8 Sept 2026Resignation statement and follow-up thread ↗
Jacob Coxon. Transcribed from screenshots supplied for this project. The original post is linked; individual reply URLs and the underlying private conversations have not been independently verified.
Opening post ↗ · Replies ↗. Supplied screenshots. Capture date unknown.
- Interview 9 Sept 2026Interview on Coxon’s departure and equity ↗
Axios. Madison Mills’s interview; employment and equity details are reported, not independently audited here.
- First-party report 8 Aug 2024GPT-4o System Card · contributor credits, page 26 ↗
OpenAI. Lists Jacob Coxon among language core contributors; does not establish employment start or end dates.
- Commentary 9 Sept 2026Coordination allegation concerning Coxon’s post ↗
Parker Thayer. Original post inspected on X. Thayer identifies himself as a Capital Research Center researcher. His inference of a coordinated campaign is an allegation.
- Interview 9 Sept 2026Coxon’s response to Bret Baier’s third-party question ↗
Peter J. Hasson / Fox News clip. Hasson’s post provides a short Special Report clip and quotes the exchange. The denial is attributed to Coxon; it is not independent verification.
- Organizational disclosure Accessed 10 Sept 2026Public grant recommendations and matching pledges ↗
Survival and Flourishing Fund. 2025 recommendations name Jaan Tallinn as funding source. The double-dagger marks conditional matching pledges, not confirmed disbursements.
- First-party report 28 May 2021Anthropic’s Series A funding announcement ↗
Anthropic. Identifies Jaan Tallinn as lead investor and Dustin Moskovitz as a participant in the 2021 round; does not establish their current holdings or board roles.
- Public record Filed 15 May 2024Good Ventures Foundation · fiscal year ending June 2023 ↗
IRS filing / ProPublica. The grants-paid schedule lists Jacob Coxon, $20,159, Long-Term Future Scholarship Program — Research Scholarship. The fiscal-year record does not supply an exact award date.
- Organizational disclosure Accessed 10 Sept 2026Newspeak House fellowship listing ↗
Newspeak House. Lists Jacob Coxon as a fellow. The listing does not document involvement in his resignation or post.
- Public record 3 Sept 2026Sanders and Casar announce proposed superintelligence restrictions ↗
Senator Bernie Sanders. An announcement of proposed legislation, published before Coxon’s resignation; not an enacted law.
- Interview 9 Sept 2026Interview with Coxon and reactions to his warning ↗
TIME / AOL. Harry Booth’s TIME interview, syndicated by AOL. Includes Coxon’s stated plans and Evan Hubinger’s personal response.
- Interview 10 Sept 2026Extended interview with Jacob Coxon ↗
CBS News. Original statement.