Three fired OpenAI safety researchers say they were never told what they actually leaked
decoder ·
Three former OpenAI safety researchers, Tomek Korbak, Mikita Balesni, and Jasmine Wang, deny allegations that they leaked confidential information, asserting they were never explicitly informed of the specific data they allegedly mishandled. Their terminations followed an internal investigation triggered by suspicions that they shared details about model architecture with external safety expert Ryan Greenblatt via Signal chats. While OpenAI cited policy violations regarding sensitive information handling, the researchers argue their dismissals are linked to their advocacy for greater transparency in AI monitoring and inspection processes. The controversy centers on a technical debate over whether companies should allow outsiders to inspect the 'chain of thought' reasoning steps used by advanced models. The researchers fear these firings will discourage independent auditors from collaborating with AI firms and create a chilling effect on employees raising safety concerns internally. Additionally, recent audits have highlighted security gaps in OpenAI’s infrastructure, though the company claims to have addressed these issues and resumed training operations.
decoder