An Anthropic researcher quit, warning AI could be out of human control by the end of 2027

By Dennis Singleton · Markets Writer · September 8, 2026, 8:45 PM ET

Key points

Coxon says neither Anthropic nor OpenAI is acting responsibly
He says systems could be out of control by end of 2027
He works on pretraining, not on a safety team
Anthropic published a similar warning this year

A researcher at Anthropic said Tuesday that he is leaving the AI industry because he fears AI systems could be out of human control by the end of 2027. Jacob Coxon told The Wall Street Journal that he doesn't want to take part in an industry-wide race to build systems that can improve themselves. Coxon works on pretraining, rather than on a dedicated safety team.

"We're on track for a lot of the most aggressive of these scenarios where by the end of next year things could be out of control already," Coxon told the Journal. He said safety trade-offs are inevitable while companies race each other and Chinese rivals, according to the report, which was written by Amrith Ramkumar and published Tuesday evening.

Coxon announced the resignation himself on X at 8:04 p.m. Eastern on Tuesday, and named OpenAI alongside his employer. "I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic," he wrote. "Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives."

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. More thoughts below.
— Jacob Coxon (@hilbertspaess) September 9, 2026

Who Coxon is

Coxon is 27, British, and studied mathematics before moving into AI, the Journal reported. He said on X that he spent the last three years on pretraining research at the two companies. Pretraining is the stage where a model learns by consuming large volumes of data. He said many colleagues now use the words "crunchtime" and "endgame" to describe where self-improving models are heading. His Google Scholar page lists an Anthropic affiliation and credits him on OpenAI's GPT-4o system card in 2024 and on an OpenAI interpretability paper published last year.

Anthropic said something similar this year

In a post titled "When AI builds itself," Anthropic policy staff Marina Favaro and Jack Clark described a threshold they called recursive self-improvement, "an AI system capable of fully autonomously designing and developing its own successor." They wrote that "we are not there yet, and recursive self-improvement is not inevitable. But it could come sooner than most institutions are prepared for." The post argued for building the ability to coordinate and verify a slowdown or pause across frontier labs, rather than for a pause on its own. Anthropic said it would slow down if other developers at or near the frontier did so "in a verifiable manner."

His departure follows the February resignation of Mrinank Sharma, who led Anthropic's Safeguards Research team and publicly warned about AI risks in a letter posted to X, Semafor reported.

Anthropic is privately held. The company confidentially filed IPO paperwork with the Securities and Exchange Commission and investors have floated a $2 trillion valuation for a possible October listing, Bloomberg reported last month.

The resignation also lands in the week that Senator Bernie Sanders and Representative Greg Casar introduced a bill that would ban superintelligent AI and pause advanced development until a federal regulator writes safety rules.

Frequently asked questions

Who is Jacob Coxon?

Jacob Coxon is a researcher at Anthropic who works on pretraining, the stage where an AI model learns by consuming large volumes of data. The Wall Street Journal reported he is 27, British, and studied mathematics before moving into AI. His Google Scholar page lists an Anthropic affiliation and credits him on OpenAI's GPT-4o system card in 2024 and on an OpenAI interpretability paper published in 2025. He says on X that he spent the last three years on pretraining research at both companies, rather than on a dedicated safety team.

Why is Jacob Coxon leaving the AI industry?

He told The Wall Street Journal on September 8, 2026 that he does not want to take part in an industry-wide race to build AI systems that can improve themselves, which he worries could spiral out of control. He said the industry is on track for the most aggressive scenarios and that things could be out of control by the end of next year. He also said safety trade-offs are inevitable while companies compete with each other and with Chinese rivals. Announcing the resignation on X, he wrote that neither Anthropic nor OpenAI is acting responsibly and that they are racing straight to self-improving superintelligence and gambling with our lives.

Has Anthropic warned about self-improving AI itself?

Yes. In a post titled 'When AI builds itself,' Anthropic policy staff Marina Favaro and Jack Clark described a threshold they called recursive self-improvement, an AI system capable of fully autonomously designing and developing its own successor. They wrote that we are not there yet and that it is not inevitable, but that it could come sooner than most institutions are prepared for. The post argued for building the ability to coordinate and verify a slowdown or pause across frontier labs, saying Anthropic would slow down if other developers at or near the frontier did so in a verifiable manner.

Have other Anthropic researchers resigned over AI risk?

Coxon's departure follows the February 2026 resignation of Mrinank Sharma, who led Anthropic's Safeguards Research team and publicly warned about AI risks in a letter posted to X saying the world is in peril, according to Semafor.

More coverage

News

Cerebras (CBRS) insiders sold $29.6 million on the day the stock jumped 10%. Since June, they have sold $296 million.

Three Cerebras insiders, including the CEO and the chief technology officer, sold $29.6 million of stock on September 4, the day the shares rose 10 percent. Two of the sales were under plans set up in May. Since selling began in June, six insiders have sold $296 million, and three more lockup releases are due by the end of October.

Sep 8, 2026

News

A YouTube channel spent $70,000 testing LG TVs. They found device tracking and voice transcripts.

Gamers Nexus and Level1Techs say LG smart TVs map every device on a home network, fingerprint what is on screen and write voice transcripts into logs on the set. LG has said its TVs do not record ambient conversations.

Sep 8, 2026

News

Meta (META) launches Muse AI agent with a free tier and plans up to $100 a month

Meta released Muse on Tuesday, a personal AI agent that books, buys and fills out forms on a user's behalf. It runs in a dedicated virtual machine with a second agent, Sentinel, approving anything it sends to the internet. A free tier and $20 and $100 monthly tiers, for US adults.

Sep 8, 2026

Dennis Singleton · Markets Writer

Dennis Singleton has spent years following the markets, but what keeps his attention is how AI is built. He writes about the companies behind the technology, from semiconductor designers and advanced packaging to photonics, memory, networking, and the hardware powering modern AI. His approach starts with filings, earnings, and industry research, then translates the important details into clear, straightforward analysis without unnecessary hype.