VERDICT: TRUE **What the claim is:** The TikTok text is a **verbatim copy of CBS News's own caption** for its Sept. 10–11, 2026 video segment — see the identical wording on [CBS News: "Former Anthropic researcher Jacob Coxon says prepping won't help in AI takeover"](https://www.cbsnews.com/video/former-anthropic-researcher-jacob-coxon-says-prepping-wont-help-in-ai-takeover/). So the claim is accurately attributing what CBS News aired. Each factual element checks out against primary sources. **Claim-linked finding 1 — Coxon really is a former Anthropic researcher who resigned while warning of human-destroying AI.** A U.S. Senate press release (Sen. Ruben Gallego, Sept. 9, 2026) states: "former Anthropic researcher, Jacob Coxon, quit his role and warned of an industrywide rush to build AI systems that can improve themselves and, in turn, **destroy humanity**," adding that Anthropic scientist Evan Hubinger put the chance AI "destroys humanity" in the next decade at more than ten percent: [gallego.senate.gov](https://www.gallego.senate.gov/news/press-releases/gallego-urges-senate-leadership-to-establish-bipartisan-senate-select-committee-on-artificial-intelligence/). **Claim-linked finding 2 — his resignation and extinction warning are recorded in official/institutional sources, not just coverage.** The OECD's intergovernmental AI Incidents Monitor logs the episode: "these incidents, along with concerns over rapid AI development, led researchers like **Jacob Coxon to resign and publicly warn about insufficient safety measures**": [oecd.ai](https://oecd.ai/en/incidents/2026-09-11-afb5). A second government source, Sen. Bernie Sanders' prepared Senate remarks (Sept. 15, 2026), quotes Coxon directly: "The people building AI earnestly believe that it **could kill us all by the end of the decade**": [sanders.senate.gov](https://www.sanders.senate.gov/press-releases/prepared-remarks-sanders-regulating-ai-is-as-american-as-apple-pie/). **Claim-linked finding 3 — a university source confirms the same resignation and warning.** The UMass Amherst Poll report (Sept. 14, 2026, University of Massachusetts Amherst) states: "**Jacob Coxon, a former OpenAI and Anthropic researcher, resigned from Anthropic**, declaring that the company was not 'acting responsibly' and warning that many in the industry believe AI 'could kill us all by the end of the decade'": [umass.edu](https://www.umass.edu/poll/about/reports/2026-09-national-public-opinion-poll-2). **Claim-linked finding 4 — his research identity is independently documented.** Coxon is a co-author (with Leo Gao, Bowen Baker, Dan Mossing et al.) of a mechanistic-interpretability paper, confirming he was a working AI researcher before the resignation: [arxiv.org/abs/2511.13653](https://arxiv.org/abs/2516.13653) (corrected: [arxiv.org/abs/2511.13653](https://arxiv.org/abs/2511.13653)). **Claim-linked finding 5 — the "not prepping / it won't do much good" element.** This comes only from the CBS News interview segment itself, which is titled "says prepping won't help in AI takeover," with the caption: he "tells Jo Ling Kent that he's not prepping for his prediction because he doesn't think it will do much good" ([CBS News](https://www.cbsnews.com/video/former-anthropic-researcher-jacob-coxon-says-prepping-wont-help-in-ai-takeover/)). The correspondent, Jo Ling Kent, corroborated the topic on X, saying they discussed "why he's still holding OpenAI equity but not Anthropic + how some in tech are 'prepping for the wrong disaster'." No government, university, intergovernmental or peer-reviewed source I found independently documents that specific prepping exchange — it is sourced to the interviewer's own outlet. **Nuance worth flagging (does not change the verdict):** "predicting that AI **will** destroy all humanity" is CBS's stronger framing. In primary records, Coxon's language is probabilistic — "could kill us all by the end of the decade," with a >10% probability cited by Hubinger — and in the same interview he said today's AI is safe for everyday use and does not pose an imminent threat ([CBS News report](https://www.cbsnews.com/news/anthropic-researcher-jacob-coxon-ai-warning/)). So the claim is accurate as a description of what CBS published about Coxon, but his own stated position is a serious risk of extinction, not a certainty that AI will destroy everyone. **Primary sources cited (all opened):** 1. U.S. Senate (government): https://www.gallego.senate.gov/news/press-releases/gallego-urges-senate-leadership-to-establish-bipartisan-senate-select-committee-on-artificial-intelligence/ 2. U.S. Senate (government): https://www.sanders.senate.gov/press-releases/prepared-remarks-sanders-regulating-ai-is-as-american-as-apple-pie/ 3. OECD.AI (intergovernmental): https://oecd.ai/en/incidents/2026-09-11-afb5 4. UMass Amherst Poll (university): https://www.umass.edu/poll/about/reports/2026-09-national-public-opinion-poll-2 5. arXiv (Cornell University-operated repository, paper record): https://arxiv.org/abs/2511.13653 6. Origin of the claim's exact wording (news outlet, not counted as a primary source): https://www.cbsnews.com/video/former-anthropic-researcher-jacob-coxon-says-prepping-wont-help-in-ai-takeover/