OpenAI Fires Three Safety Researchers Over Contacts With Outside Organizations

Three researchers who until recently worked on AI safety at OpenAI say they were abruptly fired last week. In their account, the reason was their communication with outside organizations working on AI safety. OpenAI confirmed the dismissals themselves but described the motives differently. For now, the story has two versions and very few verifiable details.

What we know so far

The original report is short, and it is important not to speculate on the participants' behalf. It does not name the fired researchers or the organizations, and it does not describe the content of the contacts. OpenAI's wording is not retold in detail either: all that is known is that it differs from the researchers' version.

• Number fired: three AI safety researchers.
• Timing: last week; the dismissal is described as abrupt.
• The fired researchers' version: the trigger was contacts with outside AI safety organizations.
• OpenAI's position: the dismissals are confirmed, but the reason is explained differently.
• The main fear the fired researchers cite: employees will become afraid to discuss AI risks openly.

The key question is what exactly the researchers shared with outsiders. Discussing public work, taking part in conferences and contributing to independent evaluations fall into one category. Passing along internal data about models that have not yet been released falls into a completely different one. The answer determines whether the story looks like punishment for talking about risks or like a breach of confidentiality.

Why this story hits a sore spot for OpenAI

The company already has a history of conflicts over freedom of expression. In May 2024, Jan Leike, co-lead of the Superalignment team, left OpenAI. He said that safety culture had taken a back seat to shiny products. It then emerged that the company's exit agreements prohibited criticism of it and tied silence to the right to already-earned equity. Sam Altman admitted the mistake and promised to fix the terms. In June 2024, current and former employees of OpenAI and Google DeepMind published an open letter on a "right to warn" about AI risks.

Against that background, the new dismissals read differently from an ordinary personnel story. Even if OpenAI has formal grounds, the timing and the makeup of those fired will draw attention. Safety researchers serve as the industry's channel to the outside expert community. Labs themselves bring in independent evaluators and testers, and the line between acceptable collaboration and a leak needs to be spelled out in advance and in clear terms.

There is also a legal layer. In 2025, California passed the law SB 53, which includes provisions protecting employees of AI developers who report catastrophic risks. Whether this case falls under them is not yet clear, and lawyers will most likely have to settle the question.

What to watch next

Until the details come out, it makes sense to read this story through a few markers. First: will OpenAI name a specific reason and cite a breach of its confidentiality policy? Second: will the fired researchers publish a timeline and documents? Third: will the safety organizations themselves speak about their contacts? Fourth: will lawsuits or appeals to regulators follow?

Employees of AI companies would do well to reread their contracts and disclosure policies now and understand clearly where the line between public and internal information runs. Corporate customers building products on OpenAI models should ask their vendors how channels for reporting risks are set up and whether the people who use them are protected. For the market, this is as much a sign of a company's maturity as the quality of the model itself.