What changed
Former OpenAI researchers Mikita Balesni, Tomek Korbak and Jasmine Wang say they were dismissed for raising safety concerns, not for mishandling sensitive information.
They published a joint open letter alleging retaliation. OpenAI says its investigation found a “significant breach of trust,” and that the dismissals were unrelated to safety criticism. The company said on 2 October that the researchers had mishandled sensitive information and said it is finalising contracts with third-party safety assessors.
Why it matters
The dispute puts a practical question at the centre of AI safety: can researchers challenge company decisions without risking their jobs?
Korbak says he had warned that OpenAI was losing the ability to monitor what AI agents “think,” which he described as an important way to catch misbehaviour. That allegation has not been independently established. But the disagreement itself matters because both sides are describing the same dismissals in opposite terms: the researchers see punishment for speaking up; OpenAI sees a serious breach of trust.
That difference affects more than three former employees. If safety researchers become reluctant to raise concerns, internal warnings may arrive later or become less candid. If OpenAI’s account is correct, the company is defending a boundary around confidential information while insisting that safety criticism remains protected.
The next meaningful test will be the promised third-party safety assessments. Their details, due in the coming weeks, should show how much scrutiny OpenAI is willing to place outside its own internal process.
Comments
No comments yet.