Three safety researchers fired by OpenAI are publicly disputing the company’s account of alleged policy violations. Tomek Korbak, Mikita Balesni and Jasmine Wang say they did not share confidential information and were never given a written explanation identifying what material they supposedly mishandled.
OpenAI’s security team investigated possible disclosures about internal model architecture and examined Signal conversations on personal phones, according to reporting summarized by The Decoder. The researchers used disappearing messages in chats with outside safety researchers. OpenAI said Signal was not an approved tool but maintained that its use was not the reason for the dismissals.
Wang faced a separate issue involving access to an executive’s inbox. She says she had previously asked for that access to be removed, accidentally opened one sensitive email and immediately reported it. OpenAI says the access itself was not the issue.
The researchers believe their advocacy for models that are easier to scrutinize may have contributed to the conflict; OpenAI denies that safety concerns they raised played a role. With the central evidence not public, neither account can be independently resolved. The three are asking OpenAI’s board for closer cooperation with external safety organizations and warn that the case may discourage staff from reporting concerns.