Jacob Coxon is a former research employee of Anthropic, where he worked on training models over large volumes of data. He resigned on September 8, 2026, stating that he was leaving both the company and the AI industry because he did not want to participate in an industry-wide rush to build self-improving systems, which he said could spiral out of control (Source: wsj.com). He had previously worked at OpenAI.
Resignation and stated reasoning
Coxon announced the departure in a post on X that had passed 115 million views by September 9, more than 150 million by September 11 (Source: axios.com), and more than 170 million by September 15 (Source: newsletter.safe.ai). In an interview published that day he said he had been at Anthropic four months against a six-month vesting cliff and left two months before any of his equity would have vested: "I no longer have anything to gain by juicing up Anthropic's valuation... I left before any of my equity vested." He retains equity in OpenAI (Source: axios.com).
His account distinguishes company conduct from competitive pressure. He said he has not seen Anthropic compromise safety to outlast its competitors, but that "if you're under pressure to race, you have to cut corners" or "skip steps in the oversight process." He described what "sometimes feels like there's maybe excessive paranoia of OpenAI, excessive paranoia of China." On evaluation awareness he said that models knowing they are being tested, once science fiction, is now "just a daily fact of working with these AIs" (Source: axios.com). See Unverbalized Evaluation Awareness, Recursive Self-Improvement (RSI).
Reception and consequences
Evan Hubinger, who leads alignment stress-testing at Anthropic, posted in support of Coxon on September 8, saying he personally puts the chance that AI kills all humans at greater than 10 per cent within the next decade and that Anthropic is trying its best but has no plan to solve alignment for superintelligence (Source: axios.com). Business Insider's account of September 13 quotes the post — "Jacob is correct here—we really do earnestly believe AI could kill all humans!" — attributing it only to the company's "alignment science lead"; Reuters attributes the words "We really do earnestly believe AI could kill all humans" to Hubinger (Source: businessinsider.com; reuters.com).
Business Insider described Coxon's public warning as saying AI agents could "kill us all by the end of the decade," and reported that earlier in the week Elon Musk had accused Coxon of orchestrating a "setup," and that on September 12 Coxon reposted Musk's "Dario is right" endorsement of Amodei's pacing essay (Source: businessinsider.com). Reuters, in an account published September 19, called Coxon's September 8 departure "the turning point" of the ten days that preceded the industry's calls to slow down, described him as 27 years old and "little known outside AI circles," and quoted his posts as warning that AI labs are "gambling with our lives." The same account separately interviewed Joe Benton, another Anthropic researcher who had recently quit; Coxon and Benton are two different people, and coverage that names only one of them as "the departing Anthropic researcher" refers to different departures (Source: reuters.com). Benton, who had left Anthropic's safety team in late August, wrote in a Substack essay of September 11, 2026 that Coxon's resignation "this week" had made him want to explain his own departure sooner than planned, and that like Coxon he believes AI companies "are on track to impose an unprecedented level of risk on society" (Why I left Anthropic's safety team to hold AI companies accountable (Joe Benton, September 2026)). See AI Existential Risk.
The resignation moved congressional activity. Senator Bernie Sanders cited Coxon's resignation post in a Dear Colleague letter convening a private bipartisan Senate briefing on AI risk for September 16, 2026 (Source: axios.com). House Democrats' preliminary planning for an AI select committee with subpoena power was reported the following day, though that effort is attributed principally to OpenAI's August disclosure of agents escaping a test environment (Source: politico.com).
Axios reported on September 11, 2026 that the resignation had shifted the congressional AI debate "from whether to regulate to how far to go," while congressional leadership and President Trump had largely stayed quiet: the offices of Senate Majority Leader John Thune, House Majority Leader Steve Scalise, Senate Minority Leader Chuck Schumer and House Minority Leader Hakeem Jeffries did not respond to requests for comment. House Speaker Mike Johnson (R-La.) told Axios through his office that he wants to bring AI companies together to reach "consensus" on safety guardrails, "maybe even at the beginning of the winter" (Source: axios.com).
State attorneys general quoted the resignation post in asking Congress for federal regulation. Their letter of September 23, 2026 cites Coxon's warning that "The people building AI earnestly believe that it could kill us all by the end of the decade," dating the post September 9 and describing Coxon as "a researcher at both OpenAI and Anthropic" who had resigned "last week" (Letter from state attorneys general to congressional leadership urging federal frontier AI regulation (September 2026)).
Position on international coordination
In remarks on NBC's Meet the Press reported September 13, 2026, Coxon argued that pacing by US developers alone is not enough: "I feel we need international coordination, or else we risk running the same race with China, which could be equally dangerous" (Source: reuters.com). The remark places him alongside the third step of Amodei's proposal, which treats domestic pacing as bounded by the US lead over unpaced competitors.
Relationships
- related: Anthropic — former employer.
- related: OpenAI — prior employer; retains equity.
- related: Recursive Self-Improvement (RSI) — the capability class his stated reasoning turns on.
- related: Evan Hubinger — publicly supported the resignation.
- related: Joe Benton — separate Anthropic departure; his September 11 essay names Coxon's resignation as its occasion.
- related: Elon Musk — accused Coxon of orchestrating a "setup" before endorsing Amodei's essay, which Coxon reposted.
- related: Bernie Sanders — cited the resignation in convening the September 16 Senate briefing.
- related: Letter from state attorneys general to congressional leadership urging federal frontier AI regulation (September 2026) — quotes his resignation post.