OpenAI Suspected Signal Leaks to Greenblatt, WSJ Reports

OpenAI suspected three of its own safety researchers of leaking internal model designs to AI safety researcher Ryan Greenblatt, including through disappearing Signal chats, according to a Wall Street Journal report on October 22.

The three, Jasmine Wang, Tomek Korbak and Mikita Balesni, were fired on October 1. An OpenAI spokesperson said an internal investigation confirmed the researchers had 'mishandled sensitive information outside established company procedures,' which the company has elsewhere called a 'significant breach of trust.'

The three dispute the account. Korbak, who served as OpenAI's technical point of contact for the outside evaluator METR, said he was told verbally he was fired over how he communicated with that group. 'Talking to METR was my job,' he said. Balesni said he was fired for 'speaking too much to third-party safety organizations.' Wang, a program manager, said the company cited her access to an executive's email, which she says had been delegated to her for recruiting; when she accidentally opened a sensitive message, she told the executive within minutes.

Days after their dismissals, the three wrote to OpenAI's board urging the company and its rivals to stop building models whose reasoning humans can no longer audit. 'As an industry, we do not yet know how to safely develop and deploy models that we cannot monitor,' the letter said, warning that the abrupt firings had created a 'chilling' atmosphere for employees still working on alignment. Korbak and Balesni were lead authors of the 'Chain of Thought Monitorability' paper that makes the technical case for keeping frontier models' step-by-step reasoning legible to human reviewers.

OpenAI did not publicly disclose the specific communications it says triggered the firings.