Three fired OpenAI safety researchers deny leaking confidential information
Former OpenAI safety researchers Tomek Korbak, Mikita Balesni and Jasmine Wang deny sharing confidential information and say the company never told them in writing what they allegedly mishandled.
Three former OpenAI safety researchers who were dismissed by the company are publicly disputing the allegations behind their firing. Tomek Korbak, Mikita Balesni and Jasmine Wang deny having shared confidential information, and they say OpenAI never identified in writing which material they were supposed to have mishandled.
The researchers also link the dispute to their advocacy for AI systems that can be more easily scrutinized. In their account, the question of what they supposedly disclosed has been left unresolved, leaving them without a clear description of the conduct the company objected to. Their public statements shift the focus from the dismissals themselves to the process used to reach them.
The dispute raises a broader question about how AI companies handle internal information and the people who work on safety. Researchers who argue for transparency and external evaluation of advanced systems often depend on access to internal findings, which makes the boundary between legitimate concern and a confidentiality breach a sensitive matter. Because the three say they never received a written statement of the specific information at issue, the exact nature of the allegation remains unclear from the public record.
The available account comes from the researchers themselves, and it describes a conflict over documentation rather than a verified finding about what was disclosed. Without a written list of the material in question, neither the researchers nor outside observers can assess the substance of the claim. The researchers' position is that they cannot respond fully to an accusation they have never seen defined.
For the wider AI sector, the case highlights the tension between corporate confidentiality and the demand for independent scrutiny of safety work. Safety teams at frontier AI developers are frequently cited as an important check on how powerful models are built and deployed, and the treatment of staff who speak publicly about those concerns is closely watched. The researchers' insistence that they were never told what they allegedly leaked puts the process itself under scrutiny.
4
