A Former OpenAI Researcher Just Quit Anthropic Saying Both Companies Could Kill Us All
An insider warning from someone who built the technology is rattling people who had never thought twice about AI safety. Here is what he actually said, and why it matters.
Key points
- A former OpenAI researcher resigned from Anthropic in May 2025, warning that neither company is acting responsibly on AI safety.
- The researcher stated that people actively building AI "earnestly believe it could kill us all by the end of the decade".
- The resignation letter went viral, suggesting millions of ordinary people are confronting AI risk seriously for the first time.
- Anthropic was founded specifically to focus on AI safety, making the criticism from an insider especially striking.
What did he actually say?
The researcher did not mince words. He said the people building AI, right now, in real offices, genuinely believe the technology could end human life within a decade. Then he said they are building it anyway.
That is the part that stopped people scrolling.
This was not a protestor outside a conference. This was someone with a desk inside one of the two most prominent AI laboratories on the planet. He worked at OpenAI, the company behind ChatGPT, then moved to Anthropic, the company behind the Claude family of AI assistants. Anthropic was created partly as a breakaway from OpenAI by people who felt safety was not being taken seriously enough. That makes his resignation letter a very uncomfortable document for both organisations.
His exact words: "This is not a marketing stunt."
Should ordinary people be worried?
Yes, paying attention is reasonable. No, panic is not the instruction here.
For years, AI safety researchers have argued behind closed doors, in academic papers, and at conferences that advanced AI systems, meaning software that can reason, plan, and act across long chains of steps without human guidance, carry risks that nobody fully understands yet. What is new is that someone with genuine insider access has said it plainly, in public, and walked away from a well-paid job to do it.
The Guardian AI reporter Garrison Lovely, who has covered AI risk for years, noted the story was not news to him. What surprised him was how many ordinary people seemed to be hearing it for the first time.
That gap matters. Decisions about how fast to build AI, what guardrails to put on it, and who gets to decide are largely happening in boardrooms and government committees, not in public squares. A nurse, a teacher, or a shop owner has as much stake in those decisions as any venture capitalist, but far less of a seat at the table.
What happens next?
Neither OpenAI nor Anthropic has, as of publication, responded substantively to the resignation.
The pressure landing on AI labs from insiders is a pattern worth watching. Researchers who leave and speak publicly are one of the few reliable windows into what is happening inside organisations that treat most of their work as confidential. Each resignation that makes headlines increases the chance that regulators, and eventually lawmakers, feel compelled to act.
For now, the most useful thing anyone can do is stay informed. Understand that when companies talk about "alignment", they mean the unsolved problem of making sure an AI system actually does what humans want rather than finding a shortcut that technically satisfies its instructions while ignoring the spirit of them. That problem is real. The people who built the tools admit it.
That admission, from inside the machine, is what made this week different.
