Fired OpenAI Pair Balesni, Korbak Dispute 'Breach' Line on NPR

OpenAI fired three safety researchers on October 1 and called it a 'significant breach of trust.' Two of them, Mikita Balesni and Tomek Korbak, now say the official line does not match what the company told them in private.

In their first on-the-record interview since the firing, on NPR's Machine Gods podcast hosted by Kevin Roose and Casey Newton, the pair said the real friction was over how safety researchers inside the company talk to outside auditors. Korbak served as OpenAI's technical point of contact for METR, the independent nonprofit evaluator. 'Talking to METR was my job,' he said. Balesni said he was fired for 'speaking too much to third-party safety organizations.'

The Wall Street Journal has reported OpenAI suspected the three of leaking internal model designs to outside safety researcher Ryan Greenblatt via disappearing Signal messages. In its own statement, reported by The Hacker News, the company said an internal investigation confirmed they 'mishandled sensitive information outside established company procedures, violating our policies and breaking the trust essential to our work.'

Korbak and Balesni are lead authors of the 'Chain of Thought Monitorability' paper, which argues frontier models' step-by-step reasoning should remain readable to human reviewers. In a joint letter after the firing, the three wrote: 'We do not believe the path to superintelligence can be navigated safely if the people closest to the risks can no longer work in high-trust, high-bandwidth ways with each other and with third parties.'

Korbak's own read on the mood before all this: 'I felt like OpenAI was broadly acting in good faith.' Balesni, asked about independent audits of frontier labs today: 'It's not going very well in our opinion.' One more entry in a run of 400 OpenAI stories we've logged in the last 90 days.