“Jacob Coxon resigned from Anthropic after spending three years doing pretraining research at both OpenAI and Anthropic, stating that neither company is acting responsibly and that they are racing toward self-improving superintelligence while 'gambling with our lives.'”
Plain restatementA researcher named Jacob Coxon publicly announced his resignation from Anthropic, described three years of pretraining research work at OpenAI and Anthropic, and stated that neither company is acting responsibly and that they are racing to self-improving superintelligence and gambling with our lives.
This one checks out. Jacob Coxon really did resign from Anthropic on September 8, 2026, and the quote in the post is word for word what he wrote on X. Major outlets including TIME, Axios, TechCrunch and CBS News confirmed the resignation and interviewed him directly, and Anthropic issued a response. The other quotes shown in the slides also match his original thread, and the caption's claim that Anthropic's alignment lead Evan Hubinger put the odds of AI causing human extinction within a decade at over 10 percent is accurate as his stated personal view. Two things the post leaves out: most of those three years were at OpenAI, not Anthropic, where he worked only a few months in 2026, and his job was pretraining, meaning building AI capability rather than safety testing. Also worth remembering that his claims about what future AI will be able to do are predictions and personal opinions, not proven findings. The post accurately reports what he said. It does not establish that what he said about the future is correct.
Jacob Coxon [drifted from the evidence:] resigned from Anthropic [drifted from the evidence:] after spending three years [drifted from the evidence:] doing pretraining research at [drifted from the evidence:] both OpenAI and Anthropic, [drifted from the evidence:] stating that neither company is acting responsibly and that they are racing [drifted from the evidence:] toward self-improving superintelligence [drifted from the evidence:] while 'gambling with our lives.'
[added by the neutral restatement:] A researcher named Jacob Coxon [added by the neutral restatement:] publicly announced his resignation from Anthropic, [added by the neutral restatement:] described three years [added by the neutral restatement:] of pretraining research [added by the neutral restatement:] work at OpenAI and Anthropic, [added by the neutral restatement:] and stated that neither company is acting responsibly and that they are racing [added by the neutral restatement:] to self-improving superintelligence [added by the neutral restatement:] and gambling with our lives.
Red-tinted words in the claim drifted from the evidence. Green-tinted words are what a neutral restatement needs.
The trace / claim to source
- The resignation happened and is confirmed by multiple independent outlets that interviewed Coxon directly.
- The headline quote is verbatim, not paraphrased or cropped in a meaning-changing way.
- The "three years doing pretraining research at both OpenAI and Anthropic" phrasing is his own exact wording.
- The quotes reproduced on slides 3, 4 and 5 match the original thread.
- The caption's summary of his argument about hacking, transforming industries, and acquiring real-world resources tracks his actual wording.
- The Evan Hubinger greater-than-10 percent extinction estimate is real, is his stated personal view, and was posted in response to Coxon.
- The Hugging Face incident referenced in slide 5 is a real, documented event.
- Subgroup/timeline ambiguity: the cover slide frames this as an "Anthropic researcher" who quit, and the three-year figure sits next to it. A reasonable reader could infer three years at Anthropic. In fact the large majority of that period was at OpenAI, with only months at Anthropic in 2026. This is Coxon's own phrasing, so it is not a distortion introduced by the post, but the compression removes a material detail.
- Authority framing: the post presents Coxon as a safety insider issuing a verdict on company conduct. His role was pretraining research, meaning capability development, not alignment or safety evaluation. The post does not misstate his role, but the visual framing invites readers to treat his claims as institutional safety findings rather than one researcher's opinion.
- Omitted counter-context: the post includes no Anthropic response, no critics, and no note that his forward-looking claims about superintelligence are predictions and personal assessments, not findings or evidence. Coxon's statements about what AI will be able to do are forecasts, not verified results.
- Unnecessary hedging in the caption: "has reportedly resigned" understates the evidence. The resignation is first-party confirmed and not in dispute.
- The substance of his predictions. Whether AI systems will become capable of the things he describes, and on what timeline, is contested forecasting, not something this investigation can verify. The verdict here covers whether he said these things, not whether they are correct.
- Exact tenure dates at Anthropic. Reporting says "earlier in 2026" but a precise start date was not located in a first-party source.
- Whether the Instagram slide screenshots are unaltered images of the original posts. The wording matches the original thread as reproduced by TechCrunch, Common Dreams and Deadline, so the text is accurate regardless, but pixel-level image authenticity was not independently verified.
The quoted resignation statement is verbatim and traceable to a first-party post. Coxon posted: "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." The event itself is confirmed by multiple independent outlets that interviewed him directly. TechCrunch reported that Coxon, a researcher who said he spent the last three years working on pretraining research at both OpenAI and Anthropic, accused the firms of failing to act responsibly. TIME reported that for three years Coxon helped train increasingly powerful AI systems at OpenAI and Anthropic, and that on Sept. 8 he walked away. The additional quoted slides in the post also match the original thread. Coxon wrote that these will soon be superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources, that progress is not slowing, and that the people building AI earnestly believe it could kill us all by the end of the decade. He also wrote that accepting this race and entering the "endgame" is a hubristic gamble that should not be launched from a private company's Slack. TechCrunch reproduced his line that he is optimistic about the potential for coordination and that warning shots like the Hugging Face attack have made pacing agreements between U.S. labs more viable. The Hubinger claim in the caption also checks out. Evan Hubinger, Anthropic's Alignment Science lead, wrote on X: "we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade." Fox LA reported this greater than 10% estimate was shared following the resignation of colleague Jacob Coxon.
Complete reasoning
The reply receipt is formatted for pasting into the thread where the claim is circulating.
Compact share page: verify.trueseeker.com/s/8ac03e4907ad/98Z88-TtsN_hlSBiiAAjMFd