Tomek Korbak, one of three OpenAI safety researchers who were fired, says the reason he was given was a "breach of trust."
"I was told that there was a breach of trust because I interacted with a third party through unsanctioned channels," Korbak told The Wall Street Journal. The quote appears in a story by Maxwell Zeff, Keach Hagey and Berber Jin. Safety researchers are the people who study how to keep AI systems from behaving in harmful or unpredictable ways, so a dismissal of three at once is notable on its own.
The stated rationale is the interesting part. Korbak describes a dispute about who he talked to and how, not about the quality of his work. That puts it in a different category from an ordinary performance exit. It also means the question is less about any one researcher and more about what counts as a sanctioned channel at a lab whose safety staff are, by the nature of the job, expected to raise concerns.
OpenAI has been here before in a broader sense. In 2024, its superalignment team, which worked on controlling future advanced systems, broke up after a series of departures. One of those who left, Jan Leike, said publicly that safety work had taken a back seat to shipping products. Korbak's account is a separate episode, but it adds to a pattern of friction between the company and the people it hired to check its work.
For anyone who relies on AI labs to police themselves, the practical issue is simple. If a researcher can lose a job over contact with an outsider, the rules governing that contact matter as much as any published safety policy.