OpenAI fires three safety researchers after an internal probe finds evidence they used Signal messaging—reportedly with disappearing messages—to communicate with an outside AI-safety researcher, raising concerns about leaks of sensitive information. OpenAI dismisses the researchers on October 1 for what it describes as a significant breach of trust, and says the matter involves mishandling sensitive information.
Reporting describes names including Jasmine Wang, Tomek Korbak, and Mikita Balesni, and an outside researcher, Ryan Greenblatt, as the recipient of the suspected communications. One outlet adds that OpenAI’s security team asks the researchers to unlock personal phones and provide access to Signal chats during meetings in September, before the firings.
The researchers dispute OpenAI’s account. They say they did not violate rules and instead were seeking more open collaboration across the AI-safety community. Coverage also places the event within a broader backdrop of earlier OpenAI personnel actions tied to disputes about openness versus tighter control of internal technical information.
