“Good People Refuse to Do Bad Things”: Anthropic Researcher Jacob Coxon Resigns

Jacob Coxon, who spent three years on pretraining research at OpenAI and Anthropic, announced his resignation from Anthropic on 9 September 2026 in a post on X. "Neither company is acting responsibly," he wrote. "They are racing straight to self-improving superintelligence and gambling with our lives." The post drew more than 115 million views.

What he actually said

Coxon is not a chief executive or a famous scientist: he has no PhD and joined Anthropic at 27. He left two months before his equity vested. "I no longer have anything to gain by juicing up Anthropic's valuation," he wrote. "I left before any of my equity vested."

He does not call the company evil — he told the Wall Street Journal its safety work is sincere. His accusation is colder: "if you're under pressure to race, you have to cut corners." In a follow-up post: "The people building AI earnestly believe that it could kill us all by the end of the decade. This is not a marketing stunt."

Colleagues did not contradict him

Evan Hubinger, an alignment team lead at Anthropic, agreed publicly: "Jacob is correct here — we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade." Anthropic is trying its best, he added, but the lab still has no plan to solve alignment for superintelligence.

A pattern, not an accident

Five years earlier Frances Haugen read Facebook's internal research, resigned and testified before the US Senate. The Wall Street Journal's "Facebook Files" showed slides saying the platform makes "body image issues worse for one in three teen girls", and that 13% of British and 6% of American teens reporting suicidal thoughts traced the desire to die to Instagram. In both cases the company knew — and someone inside had to decide whether knowing was enough.

The culture behind it

The problem, the author argues, is wider than a few bad companies: engineers are asked to treat quality as a luxury. A 2026 survey of tech workers named the condition "smiling exhaustion"; one respondent put it as "I can do more, faster, but not better". More than half (53%) would steer a newcomer away from their field, and a separate report found 67% of developers now spend more time debugging AI-generated code.

His closing question: if the answer is the salary, say it louder; if it is that someone worse would take the job, say that too; if it is that you can change things from the inside, show what you changed. Those who refuse are not heroes but human.