The Resignation Letters

On Tuesday 8 September 2026 Jacob Coxon resigned from Anthropic and said so in public. "I resigned from Anthropic today," his post begins. "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 three years are across both companies together, by his own account. The post is the first of a thread, and only the first is held here. By NBC's count it had been viewed more than 155 million times two days later.

The same NBC report, of 10 September, carried two more researchers: Joe Benton, who led a safety research team at Anthropic, and Josh Engels, who worked on AI safety research at Google. Both are joining METR. NBC does not say when they left, only that it was recently. Benton's reason, in his words: "I left because I think I can have more positive influence ... from outside these companies". Engels's: "There are no adults in the room". The line is his, said to NBC, and belongs to neither Benton nor Coxon.

The earliest departure in this account is from February 2024, and two more came that May. In May 2024 Ilya Sutskever, OpenAI's chief scientist, and Jan Leike, who together led its superalignment team, resigned, as Vox reported. Vox gives no reason in Sutskever's own words, and none is held here. Leike gave one in public, in a post of 17 May 2024 that sits in the middle of his own thread and begins with "But": "safety culture and processes have taken a backseat to shiny products". Four months later William Saunders, who had worked on alignment at OpenAI, told the Senate that the team "no longer exists; its leaders and many key researchers resigned after struggling to get the resources they needed".

Daniel Kokotajlo quit OpenAI, he wrote, "due to losing confidence that it would behave responsibly around the time of AGI". He left without signing the exit agreement. Vox reported on 17 and 18 May 2024 that under that agreement departing staff who would not sign, or who later broke it, could find that "they can lose all vested equity", and that Kokotajlo "had to surrender what would have likely turned out to be a huge sum of money in order to quit without signing the document". OpenAI's reply to Vox was that it has never cancelled any current or former employee's vested equity and will not, and within the week it said that none had been or would be.

On 4 June 2024 a letter asking for a right to warn about advanced AI carried 13 signatories, counted from the letter's own page. Seven signed by name and six without. Eleven were current or former staff of OpenAI and two of Google DeepMind. Four of the six who withheld their names were still at OpenAI when they signed, and all four of OpenAI's current staff on the letter were among the anonymous. Three well-known researchers endorsed it, Yoshua Bengio, Geoffrey Hinton and Stuart Russell; they are not signatories and are not in the thirteen. The pattern in the count is the letter's own argument: the people with the most to lose by warning were the ones who could not sign their names.

On 17 September 2024 William Saunders told a Senate subcommittee, under oath: "I resigned from OpenAI because I lost faith that by themselves they will make responsible decisions about AGI." The sentence is in his written testimony, and the printed record of the hearing shows the witnesses sworn and has him saying it aloud that day. He had worked at OpenAI for three years, and by the chair's introduction had resigned in February 2024, which makes his the earliest leaving in this account, three months before Sutskever and Leike. A post can be deleted and a letter can be signed without a name. Testimony is the one form here that carries a penalty for saying what is not true.

Miles Brundage, OpenAI's senior advisor for AGI Readiness, announced on 24 October 2024 that he was leaving; his team was disbanded and its members reassigned. He did not say he left in protest. By CNBC's report he said his research would have more impact outside, and that he wanted to be less biased. He also said that neither OpenAI nor any other frontier lab is ready, and nor is the world. Steven Adler, a safety researcher, left OpenAI in late 2024 after four years and said so in public on 27 January 2025. He wrote that he was "pretty terrified by the pace of AI development", and: "An AGI race is a very risky gamble, with huge downside." He left in one year and spoke in the next, which is one reason these departures do not sit well in a count by year.

The reasons are not all one reason. In February 2026 Zoë Hitzig left OpenAI over advertising. "I resigned from OpenAI on Monday," she posted on 11 February. "The same day, they started testing ads in ChatGPT. OpenAI has the most detailed record of private human thought ever assembled. Can we trust them to resist the tidal forces pushing them to abuse it?" That month the BBC reported that Mrinank Sharma, who led a team researching AI safeguards at Anthropic, had resigned with a letter saying the "world is in peril", a peril he described as "not just from AI, or bioweapons, but from a whole series of interconnected crises".

In posts, in a signed letter, and under oath, people who worked at these labs have left and said why. They are not counted here, because they are not one kind of thing. Some feared catastrophe, one feared advertising, one objected to how a defence agreement was made, some said they could do more from outside, and some who left gave no reason at all. A number would add them together as if they agreed. Read each in the words of the person who gave it.