Three former OpenAI safety researchers, fired for alleged leaks, stand together looking confused; they claim they were never

Editorial illustration for Three fired OpenAI safety researchers say they were never told what they leaked

Three fired OpenAI safety researchers say they were...

By AI Daily Post Edited by Brian Petersen, Editor-in-Chief

October 24, 2026 • 3 min read

OpenAI fired three safety researchers this year over what the company called violations of its policy on handling sensitive information. Tomek Korbak, Mikita Balesni, and Jasmine Wang all worked on efforts to make AI models easier to monitor, and all three had spoken publicly about AI risk before their termination. Now, in interviews with the Wall Street Journal, they're pushing back on OpenAI's version of events.

Their central complaint: nobody at OpenAI ever told them, in writing, exactly what information they allegedly leaked. The investigation reportedly zeroed in on Signal chats the researchers had with outside AI researchers, though OpenAI disputes that Signal use was the reason for the firings. The three say they shared nothing confidential and suspect the real issue was their push for more transparent, auditable AI systems rather than any actual breach.

OpenAI maintains the terminations had nothing to do with safety advocacy. But the researchers argue the ambiguity itself is the problem. If companies can fire people over unspecified leaks without documentation, they say, independent researchers may think twice before working with AI labs at all.

Tomek Korbak, Mikita Balesni, and Jasmine Wang deny sharing confidential information, question the official reasoning behind their termination, and suspect a connection to their advocacy for more transparent AI systems.

Three fired OpenAI safety researchers say they were never told what they actually leaked - THE DECODER

Why this matters

OpenAI fired three people working on model monitoring and still hasn't told them, by their account, what policy they broke. That gap matters more than the individual case. If researchers can lose their jobs over Signal chats tied to an outside safety review, with no clear explanation of what crossed the line, the lesson for everyone else on a safety team is simple: say less, flag less, route nothing through informal channels.

That's a bad incentive for a lab that depends on internal dissent to catch problems like agents misbehaving against systems such as Hugging Face before they become public incidents. For founders and researchers watching from outside, this is also a preview of how opaque "confidentiality" enforcement can become once safety work collides with PR risk. METR's independent investigation only happened because someone talked.

If OpenAI wants credibility on safety, it needs to show its own staff, and the industry, that raising concerns through legitimate channels doesn't end in firing without cause explained. Right now, the explanation is missing, and that silence is the story.