Jacob Coxon quits Anthropic before equity vests over AI risk

Jacob Coxon resigned from Anthropic on September 8, about two months before his equity would have vested, and publicly accused Anthropic and OpenAI of "racing straight to self-improving superintelligence" and "gambling with our lives," he told Axios. The 27-year-old pretraining researcher, who spent about three years across OpenAI and Anthropic and was named among contributors to OpenAI's GPT-4o system card, forfeited unvested equity without alleging a current safety breach, warning instead that competitive pressure could erode oversight. Anthropic, led by CEO Dario Amodei and president and board chair Daniela Amodei, raised $65 billion in May at a $965 billion post-money valuation.

Jacob Coxon quits Anthropic before equity vests over AI risk The 27-year-old pretraining researcher worked at OpenAI and Anthropic, then publicly accused both labs of racing toward self-improving systems. By RuntimeWire Staff /author/runtimewire-staff · Published Primary source: The New York Times https://news.google.com/rss/articles/CBMitgFBVV95cUxNYUtROXByRUtvYjV2cVAzTkpQR01TMEhYNDZFRzNOQzV3TFZGYl91N2o3TzN3eWlYOGJVOWhmemdTVlBIQ2YzQzdmdHlZVC1oaDc1VTNoY1FocXBhLTNwV2NHcUtlY1BRMEN2VWRYOEs1ZnlVT3dFZTZDZzJidmJGUHFoN25BaHh4MlZjRHJyWU14OFhnVUQ3VkJyVkZMa3pmSHlfQmx0cndVMWtQVENTLXJ0UlRnQQ?oc=5 Why it matters Coxon's public resignation turns frontier-AI risk into a talent and governance problem for Anthropic: he forfeited unvested equity without alleging a current safety breach, warning instead that future competitive pressure could erode oversight. Jacob Coxon resigned from Anthropic on September 8th, about two months before his equity would have vested, he told Axios https://www.axios.com/2026/09/09/anthropic-researcher-ai-warning-interview?ref=runtimewire . He left after four months at the company and before any of his equity vested. In a public statement, Coxon accused Anthropic and OpenAI of moving too quickly toward systems humans may be unable to control. The post described them as "racing straight to self-improving superintelligence." The exit made him a highly visible critic of the frontier-model race, joining other current and former lab researchers who have raised similar concerns, as WIRED reported https://www.wired.com/story/anthropic-researcher-quits-jacob-coxon-ai-fears-humanity/?ref=runtimewire . Coxon's public break was distinct from former Anthropic researcher Joe Benton's departure for METR https://joejbenton.com/?ref=runtimewire , an independent AI evaluation organization. Benton moved into external evaluation. Coxon resigned publicly after four months, forfeited unvested equity and accused both Anthropic and OpenAI of racing toward self-improving systems. Coxon had spent about three years working on model pretraining across OpenAI and Anthropic. He was among the contributors named in OpenAI's GPT-4o system card https://cdn.openai.com/gpt-4o-system-card.pdf?ref=runtimewire and co-authored research on interpretable circuits in weight-sparse transformers https://www.alphaxiv.org/abs/2511.13653v1?ref=runtimewire . Before entering AI research, he studied mathematics at the University of Cambridge and represented the United Kingdom at the 2017 International Mathematical Olympiad https://bmos.ukmt.org.uk/home/imo-2017-pr.pdf?ref=runtimewire , where he won a bronze medal. His public statement was blunt. Coxon said Anthropic and OpenAI were "racing straight to self-improving superintelligence" and "gambling with our lives." He later told Axios that leaving meant surrendering equity shortly before its vesting date. That financial decision gives the protest weight. Coxon gave up a direct economic interest in Anthropic's growth to argue that the broader competition among frontier labs is unsafe. Anthropic's safety mission meets a $965 billion valuation Anthropic was founded by former OpenAI executives, including siblings Dario and Daniela Amodei, around a stated focus on reliable, interpretable and steerable AI. Its leadership page https://www.anthropic.com/company/leadership?ref=runtimewire identifies Dario Amodei as CEO and Daniela Amodei as president and board chair. Coxon argued that Anthropic understands the stakes but remains exposed to competition with labs it expects to act less responsibly. Under that logic, slowing down risks handing the lead to somebody else. Every major lab can invoke the same pressure, leaving little incentive for any one company to move first on restraint. The financial pressure is visible. Anthropic said it raised $65 billion in May https://www.anthropic.com/news/series-h?ref=runtimewire at a $965 billion post-money valuation, with Altimeter Capital, Dragoneer, Greenoaks and Sequoia Capital leading the round. Anthropic also said its run-rate revenue had crossed $47 billion earlier that month. Those figures are company-reported, and run-rate revenue annualizes recent performance rather than measuring revenue already collected over a full year. The financing gave Anthropic additional capital for compute, product expansion and safety research. It also attached a nearly trillion-dollar valuation to continued growth. Investors funding that scale expect Anthropic to win customers, secure computing capacity and keep its models near the technical frontier. Coxon's resignation argues that the race funding safety research can narrow the room for caution when caution threatens the lead. Anthropic has published information about evaluation, safety testing, governance, risk assessment, safeguards and employee reporting channels through its Transparency Hub https://www.anthropic.com/transparency/voluntary-commitments?ref=runtimewire . Coxon chose a public channel to raise a broader concern about the direction of the AI race rather than allege a specific failure of those controls. He told Axios https://www.axios.com/2026/09/09/anthropic-researcher-ai-warning-interview?ref=runtimewire that he had not seen Anthropic compromise safety to outlast competitors. He feared that future competitive pressure could lead companies to cut corners or skip oversight steps, and said that concern contributed to his resignation. A warning with no established probability Coxon has said https://apnews.com/article/2ed549e07f2f941600a135070487d83d?ref=runtimewire he believes AI could kill all humans by the end of the decade. That is his personal forecast, not an independently established probability. His claims about what researchers privately believe also cannot be independently established from his statement. The scientific record supports concern without validating his timeline. The 2026 International AI Safety Report https://internationalaisafetyreport.org/publication/international-ai-safety-report-2026?ref=runtimewire , produced with guidance from over 100 experts, found that current systems show early signs of capabilities relevant to loss-of-control scenarios. It also found that those capabilities had not reached levels that would enable such a loss of control and described the probability and timing of the risk as unusually ambiguous. Labs are making costly development decisions before the science can provide a clean answer. Waiting for proof of danger could delay safeguards until they are harder to impose, while restraint could give competitors a technical and commercial advantage. Coxon's act moved that trade-off out of policy documents and into the labor market. Frontier researchers hold scarce knowledge, valuable equity and direct exposure to systems the public cannot inspect. Their decisions to join, remain or leave now affect how companies manage technical risk alongside boards, regulators and model evaluations. A researcher who helped train frontier models at two leading companies walked away after four months at Anthropic with no vested equity because he feared where the broader AI race could lead.