OpenAI reportedly examined researchers’ Signal chats before firing them

OpenAI reportedly asked three safety researchers to unlock their personal phones while investigating suspected disclosures about its AI model designs. Security staff are said to have found disappearing messages enabled in some Signal conversations with an outside researcher. The employees were reportedly dismissed a week later.

The dismissed researchers were Tomek Korbak, Jasmine Wang and Mikita Balesni. OpenAI has confirmed the dismissals and said its investigation established violations involving sensitive information. The company denied that it removed them for expressing safety concerns, while the researchers have challenged its explanation.

In a joint letter, the researchers denied supplying a leak about model designs that could make AI reasoning harder to inspect. They said collaboration with outside safety specialists fell within their responsibilities and warned that uncertainty over the rules could discourage such work. Korbak had served as the technical contact for METR during its investigation of an incident involving OpenAI agents and Hugging Face.

The researchers asked OpenAI’s oversight bodies to preserve independent safety auditors’ access and clarify how employees may communicate with outside organizations. They also urged the company to protect researchers’ ability to examine advanced models’ reasoning. Their letter disputes the misconduct allegations rather than resolving them.