AI 1,003 views on YouTube
He Quit Anthropic To Warn About AI. Its Safety Lead Agreed.
An Anthropic researcher resigned to warn that AI could kill everyone by the end of the decade — and the company's own alignment science lead replied by putting the number higher.
Jacob Coxon spent about three years on pretraining research, first at OpenAI and then at Anthropic. He resigned on 8 September 2026 in a seven-part post on X, writing that the labs "are racing straight to self-improving superintelligence and gambling with our lives," that "the people building AI earnestly believe that it could kill us all by the end of the decade," and that "no other human activity poses this level of danger." He told Axios he left before any of his equity vested, roughly two months short — which is what separates this from safety resignations that came after the stock landed.
Anthropic did not deny it. Evan Hubinger, the company's alignment science lead, replied on X: "I personally think it is [greater than] 10% within the next decade," and added, "I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to." Hubinger still works there.
The same week, Sen. Bernie Sanders and Rep. Greg Casar announced the Ban Artificial Superintelligence Act, and California Gov. Gavin Newsom signed two bills creating the first state AI auditing framework in the US — both backed by Anthropic and OpenAI.
The reporting
Verified 2026-09-10. Every claim below carries at least two independent sources.
Disclosure carried in the description: this reel is about Anthropic, the company that makes the model used to assemble RECORDx reels. It reports the story from named sources on both sides and takes no position on the estimate.
References
Reporting drawn from Axios · Forbes · TIME · Newsweek · Quartz · Common Dreams · sanders.senate.gov · gov.ca.gov
What the reel says
Narration as scripted and voiced.
A researcher quit Anthropic this week to warn that artificial intelligence could kill everyone. Then his own company's head of alignment answered him in public and said the odds were higher than that
His name is Jacob Coxon. He spent about three years inside this industry doing pre-training research, first at OpenAI and then at Anthropic. That is the work that actually builds the model
He resigned on the 8th of September, and he left before any of his equity vested. He told Axios he was two months short. That is what separates this from the other safety resignations because most of those came after the stock had already landed
In a seven-part post, he wrote that the labs are racing straight to self-improving superintelligence and gambling with our lives. He also wrote that the people building AI earnestly believe that it could kill us all by the end of the decade and that no other human activity poses this level of danger.
Anthropic did not deny it. Evan Hubinger, who leads alignment science at the company, replied on the same platform and put a number on it. He wrote, "I personally think it is greater than ten percent within the next decade."
He went further. He wrote that he believes Anthropic is trying its best, but that we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to
Hubinger still works there. So this is not the industry against its critics. It is one man who left and one man who stayed, agreeing on the size of the risk and disagreeing only about what to do on Monday morning
Washington moved the same week. Senator Bernie Sanders and Representative Greg Casar announced a bill to permanently ban the development of artificial super intelligence and to pause advanced AI until a federal regulator writes safety rules
In California, Governor Newsom signed two bills creating the first state auditing framework for AI in the country. Anthropic and OpenAI both backed them
Every estimate here came from someone with equity in the answer. The man who gave his up and the man who kept his and raised the number are describing the same building. Neither of them is claiming there is a plan
Disclosure: RECORDx reels are assembled with the help of a model made by Anthropic. This report is built from named public sources on both sides and takes no position on the estimate.
All footage is licensed stock. No footage of any person named here is used.
