OpenAI Just Fired 3 of Its Own Safety Researchers

OpenAI just fired the very people tasked with keeping its AI safe. On October 1, the company confirmed it "parted ways with three individuals" after an internal investigation found they had mishandled "sensitive information" outside the company's established procedures — including work with an outside organisation that evaluates AI models.

The details: According to the Wall Street Journal and Bloomberg, the three researchers were Jasmine Wang, Tomek Korbak, and Mikita Balesni — names OpenAI itself has not publicly confirmed. The exact nature of the mishandled information has not been disclosed. What is confirmed: an internal investigation found sensitive info handled outside proper procedures, and the company cut ties.

The irony: These were among OpenAI's loudest internal voices warning about AI risk. On September 10, Balesni posted on X: "i am at OpenAI and i think AI is >10% likely to kill all humans." A day later, Korbak wrote that he was unhappy with much of what OpenAI does. And Wang had warned about the dangers of recursive self-improvement, shortly after Anthropic researcher Jacob Coxon resigned. The lab that talks most about safety just dismissed the people who took safety most seriously.

The context: The timing is awkward. This same week, Nvidia, Google, Meta, xAI, OpenAI, and Anthropic all signed a voluntary White House pledge on AI safety — six frontier labs promising to be careful. Meanwhile, one of them was firing its own safety researchers.

Why it matters: Safety teams at AI labs have been shrinking across the industry — researchers leaving, being pushed out, or quietly reassigned. The people whose job is to flag risks keep exiting through the back door, exactly as the models get more powerful. The White House pledge is only as strong as the people still inside the building to uphold it.

The take: OpenAI confirmed the firings but hasn't confirmed the names or disclosed what was mishandled — and those gaps matter. Still, the pattern is hard to miss: the industry's safety infrastructure keeps thinning out at the exact moment it needs to be thickest. Everyone promised to play it safe this week. The question is who's left to keep that promise.