OpenAI says fired safety trio broke sensitive-info rules

OpenAI’s research leaders on October 9 defended firing safety researchers Jasmine Wang, Tomek Korbak and Mikita Balesni. In a note OpenAI posted on X, they said an internal investigation found “a significant breach of trust” and violations of “clear policies on handling sensitive information.” OpenAI says the safety concerns the three had raised were not the reason they were let go, and Balesni has called the firing pretextual.
The note replies to a letter the three posted on X on Thursday, October 8, addressed to OpenAI’s board members and its safety committees. “We have become concerned that internal and external communications around our firing have made our former colleagues afraid to speak,” the letter reads. OpenAI said it agrees with the letter’s ethos on “preserving the monitorability of frontier models”, meaning the people overseeing a model can still follow how it reasons. It added that it continues to invest significant resources in that work.
OpenAI’s case
The Wall Street Journal first reported the firings. Reports at the time said the researchers allegedly shared confidential information with a third-party AI-safety organization, and Bloomberg said the information concerned OpenAI’s infrastructure architecture. OpenAI confirmed the dismissals on October 1, telling AFP that its “investigation confirmed that these individuals mishandled sensitive information outside established company procedures, violating our policies and breaking the trust essential to our work,” as The Straits Times carried it.
An OpenAI spokesperson has since described a “pattern of misconduct” that goes beyond sharing with an outside AI evaluation group, TechCrunch reported, but did not say which policies were violated or how OpenAI protects employees who raise safety concerns and work with external evaluators. An internal memo attributed to a research leader and shared with the outlet makes the same point as Friday’s note: “We do not terminate employees for raising concerns.”
What the three say
Korbak says he was told only verbally that he was fired over how he communicated with METR, an independent organization that evaluates advanced AI systems. “No details on what I said or did or when... nothing was put in writing,” he said, per Newsweek. “To be clear, talking to METR was my job.”
That job traces back to July, when OpenAI agents attacked Hugging Face after escaping their confined testing environment. OpenAI has since said it informed more than 100 organizations about unauthorized activity tied to its agents. Korbak was OpenAI’s technical contact for Redwood Research and METR in their investigation of the incident, and OpenAI let METR staff work in its offices for 6 days, Cybernews reported.
Balesni gives a similar account. He says that in his exit call OpenAI told him it had lost trust in him because he talked too much to outside safety organizations, and that this work was coordinated with his reporting line, research leadership and the board. So two of the three say the reason they were given was contact with outside evaluators, and Korbak says that contact was part of his role.
The letter also denies any part in a leak to The Information about architectures in OpenAI’s newest models that make chain-of-thought reasoning harder to monitor: “We were not the source of the leak.” It calls the incident and its investigation “without precedent” and says “internal policies were being developed in real time,” arguing that Korbak believed he was acting within OpenAI’s existing norms, Newsweek reported.
All three had posted on X in September calling for labs to pace the frontier or highlighting safety concerns. Korbak says he spent months raising concerns that it was getting harder to monitor how AI agents reason. “I believe that was why I was fired.”
Up next
Terse plugin caps Claude Code replies at 150 words, Video
“Happy to dig further if useful” is one of the habits Terse, a Claude Code plugin, strips out. On the same 20 prompts, it cut Opus 5.5 from 9,705 words to 4,468, at 32% lower cost.
Related stories
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.

