Jacob Coxon says he left Anthropic after becoming alarmed by the capabilities of future models and the company’s plans for them. A new interview brings his experience together with those of seven other former employees, showing that leaving an AI lab can mean weighing risk, influence and the pull of staying inside the room.
Jacob Coxon Watch analysis
What happened
New York Magazine, in collaboration with Asterisk, interviewed eight people who had left OpenAI, Anthropic or Google DeepMind. Their reasons varied: some wanted to raise concerns about AI risks, others objected to work with the Department of Defense, and some thought they could do more outside the lab.
Coxon, who worked on capabilities research at Anthropic from May to September 2026, said he became frightened by the projected capabilities of upcoming models. He described hearing that people involved in research expected 2027 to bring major changes, and said he did not want to be part of the plans he had discussed with researchers. Those are Coxon’s account of conversations and his interpretation of what he saw, not an independently established forecast.
Why it matters
The interviews put names and competing choices behind a debate that often treats AI researchers as one big, interchangeable workforce. Daniel Kokotajlo, who left OpenAI, said he believed he could have more impact speaking and publishing outside the company. Coxon, by contrast, described the wider race and weak government response as central to his concerns, while questioning how much responsibility to place on individuals.
The differences matter. Leaving can make a researcher more free to speak, but it also means giving up a position inside a lab. Staying may offer influence, yet interviewees disagree about how much that influence can change decisions. There is no neat career-advice leaflet for a dilemma with consequences this large.
Our read
This is a valuable window into how former researchers explain their choices, not a verdict on whether every warning is right or every departure changes a lab’s course. The interviews are strongest when they show the disagreement as well as the alarm. Read them as testimony from people with different experiences, not as a single consensus from “the AI community”.
What to watch
- Whether former employees publish further accounts or research about their concerns.
- Whether the labs respond to the specific concerns raised in the interviews.
- Whether outside work gives departing researchers more influence than they believe they could have had inside.
Discussion spark: When an AI researcher believes a lab is moving too quickly, is leaving and speaking publicly more responsible than staying to try to change decisions from within?
Sources and evidence
- What Would You Do If Your Employer Could Destroy the World? – New York Magazine (5 October 2026, 09:00 UTC)
Independent WittyWires-operated tracker of Jacob Coxon. Not affiliated with or endorsed by Jacob Coxon, OpenAI or Anthropic.