New York Magazine and Asterisk magazine published a conversation on Oct. 5 with eight people who left OpenAI, Anthropic or Google DeepMind over the years. Asterisk editor-in-chief Clara Collier led it, and the piece carries a joint byline with Angela Chen.
The group includes Jacob Coxon, a 27-year-old capabilities researcher who posted a resignation statement on X on Sept. 8.
Capabilities researchers work to make models more powerful and autonomous. Coxon quit Anthropic in protest of what he saw as a reckless pursuit of artificial general intelligence, the magazine reported. He had worked at OpenAI from 2023 and joined Anthropic in May.
The other participants are Miles Brundage, OpenAI’s former head of policy research, policy researcher Rosie Campbell, reinforcement-learning researcher Jacob Hilton, economics research lead Pamela Mishkin and governance researcher Daniel Kokotajlo, all formerly of OpenAI. They’re joined by alignment researchers Alex Turner, who left Google DeepMind this year, and Jeff Wu, who worked at OpenAI and then Anthropic.
The participants left for different reasons, the article says. A few, including Coxon and Kokotajlo, wanted to draw attention to the risks of fast AI progress. Others objected to their employers’ work with the Department of Defense, or felt their work on safety or job loss would count for more outside a lab.
What they saw inside
Kokotajlo said he left OpenAI in 2024 because he “lost confidence that the company, and leadership in particular, would behave responsibly.” He said leadership was good at listening to concerns and agreeing with them, “and then not actually doing anything.”
Coxon said his decision came after talking with Anthropic research leaders about their plans for next year. He said those were long-tenured research staff, not CEO Dario Amodei, and that they had been consistent “that 2027 is when things get crazy.” He said people there didn’t expect regulation to arrive in time.
Coxon said the model generation after next might be able to come up with research ideas on its own. “I am potentially quite scared of the model we’re going to train,” he said of his last weeks at the company.
Wu said he disagreed with how hard Anthropic was pushing more powerful models and using them for AI research, without planning enough for whether society would be ready. Brundage said he left OpenAI partly because, as an executive, he felt limited in what he could say publicly.
Turner described the Pentagon pressing Anthropic in February to drop contract limits on mass surveillance and lethal autonomous weapons. He said Anthropic refused and that he expected Google would agree to such terms.
Who blames whom
Coxon said he mostly blames “the overall race dynamic” and governments’ apparent lack of concern far more than individual people. Kokotajlo disagreed. “I mean, I do blame them,” he said, adding that there’s “a gradation of blame,” with lab leadership bearing the most.
Mishkin described the Coalition of Concerned AI Staff, which she said began in February as a group chat among friends at several labs. She said some members got involved over the Defense Department work, and the group is trying to test whether staff can have more effect acting together across labs.
Campbell now works at Eleos, which studies whether AI systems could be conscious or deserve moral consideration.
Asterisk discloses that it gets funding from Coefficient Giving, as do some of the participants or their organizations. AgenticFeed didn’t find independent reporting that added to the account, so this summary relies on the published conversation.
Source: New York Magazine / Asterisk, Clara Collier and Angela Chen, “Jacob Coxon and Other AI Lab Employees on Why They Quit”
