Artificial Intelligence
Former AI lab researchers explain why they quit
The article gathered interviews with former OpenAI, Anthropic and Google DeepMind researchers about leaving their labs, focusing on safety fears, internal frustrations and disagreement over how to respond.
- On 8 September, Jacob Coxon posted a resignation statement on X and said he had left Anthropic in protest at the company’s pursuit of artificial general intelligence.
- New York Magazine and Asterisk brought together eight former employees from OpenAI, Anthropic and Google DeepMind to discuss why they had walked away.
- Daniel Kokotajlo said he left OpenAI after losing confidence in the company’s leadership and wanting more freedom to speak publicly.
- Coxon said Anthropic researchers had told him that 2027 was when they expected things to get serious, and that they did not expect regulation or international co-operation to arrive in time.
On 8 September, Jacob Coxon posted his resignation on X, and the statement spread quickly online. He had been a capabilities researcher at Anthropic and said he was leaving because he opposed what he saw as the company’s push towards AGI.
The article then assembled interviews with former employees from OpenAI, Anthropic and Google DeepMind. Their reasons for leaving included concern about risk, objections to links with the Department of Defense, and frustration that organisations built on idealism had become more corporate.
Daniel Kokotajlo said he had left OpenAI after deciding that its leaders could not be relied on to act responsibly, and that he wanted to say publicly why he had gone. Coxon said conversations inside Anthropic made him think the next model generations would be more capable and that outside regulation would come too late.
The interview also covered arguments about whether leaving a lab mattered more than staying inside to influence it. Kokotajlo said he still worked with AI-safety groups despite some discomfort, while Coxon said he blamed the wider race dynamic and the structure of the industry more than individual researchers.
Named in this story
People
- Jacob Coxon
- Anthropic capabilities researcher who resigned on 8 September
- Daniel Kokotajlo
- former OpenAI researcher interviewed about leaving the company
- Holden Karnofsky
- Anthropic staff member Coxon said he spoke to before resigning
- Nick Joseph
- Anthropic head of pretraining whom Coxon consulted
- Dario Amodei
- Anthropic leader Coxon mentioned while describing internal discussions
- Evan Hubinger
- person whose tweet Coxon said did not help Anthropic’s IPO
Companies
- Anthropic
- the AI lab Coxon left and criticised
- OpenAI
- the company Kokotajlo resigned from in 2024
- Google DeepMind
- one of the labs represented by former employees in the interview
Governments and agencies
- Department of Defense
- US defence department whose ties some interviewees objected to
How the source tells it
The piece had a highly charged, interview-led register, with repeated language of danger, blame and urgency framing the researchers’ departures.
- alarm end-of-world language and worst-case outcomes were used to describe the labs’ work
- moral framing the argument was repeatedly cast as a question of blame, responsibility and what is good for the world
- urgency near-term deadlines and crunch-time timing were used to press the sense that decisions had to be made now
- hype the field was described as the most important technology and its leading firms as exceptionally powerful