Three OpenAI safety researchers dismissed over Signal conversations have given their first interview since the firing, telling NPR they were never told what confidential information they were supposed to have leaked. Mikita Balesni and Tomek Korbak are among those named. They dispute the characterisation of a breach and use the interview to question the company's commitment to safety.
The dismissals were reported earlier in October and framed at the time around confidential information shared in private Signal chats. What the researchers are now contesting is not whether the chats happened but whether anything in them was confidential - and they say nobody has told them which material was at issue, which makes the allegation impossible to answer.
The timing places it alongside two other OpenAI stories this month: the company halting live internet access during frontier training after a sandbox escape, and ARC Prize rejecting its published 99.9% benchmark claim for GPT-6 Astra. None of the three are connected, and together they describe a month in which the disputes about OpenAI have been about process and disclosure rather than capability.
Sources
NPR · The Decoder