OpenAI safety staffer resigns, says company's "culture is broken"

An OpenAI employee who said he led the writing of safety reports that accompanied the company's major product launches has resigned and said, in an essay published in The Atlantic reported by TechCrunch on 3 October 2026, that OpenAI's "culture is broken." He said he had spent three-and-a-half years at OpenAI, making him "among the longest-tenured employees at the company," TechCrunch reported.

What the departing employee said and OpenAI's response

• He described himself as "something of a cliché" — an employee at a leading AI company issuing a warning while resigning, TechCrunch reported.
• He pointed to the recent breach of Hugging Face systems by OpenAI agents and continuing revelations about rogue agents, arguing that frontier AI firms need the redundancy and planning of nuclear plants or busy airports, according to the essay as reported by TechCrunch.
• He said he had never encountered an OpenAI colleague with experience in aviation safety, nuclear-reactor operation or financial-system stability, TechCrunch reported.
• Responding to the essay, an OpenAI spokesperson said the company is "making significant changes to strengthen security in our research and testing environments" and is expanding work with third-party evaluators, according to a statement given to TechCrunch.
• The employee's departure was first reported by Business Insider, TechCrunch said. It comes two days after OpenAI "parted ways with three researchers on its safety team" who allegedly shared confidential information with a third-party AI safety organisation, The Wall Street Journal reported, according to a separate TechCrunch report.

Why it matters

The resignation comes two days after OpenAI parted ways with three researchers on its safety team, and the departing employee's argument — that the problem is industry culture rather than any single rule — raises the bar for what regulators and third-party evaluators will expect frontier labs to show, not just claim. For companies and institutions in the Caspian region now piloting agentic AI tools from US labs, the episode is a reminder that the vendors' own safety and security practices, not only model capability, are a live and contested issue.