Three safety researchers who lost their OpenAI jobs are now describing the internal review that ended their tenure. Tomek Korbak, Mikita Balesni and Jasmine Wang deny moving confidential material anywhere, and they say the company never handed them a written account of what they supposedly mishandled.
OpenAI’s position is that an internal inquiry into its sensitive-information policies drove the decision, and that the safety criticism these three had voiced in public played no part. Company officials likewise ruled out the Signal app as the cause, while reminding staff it was never on the approved list.
Wall Street Journal coverage describes how it unfolded. Security staff suspected the group of passing architecture details to a researcher outside the company and demanded the three unlock their personal handsets. Agents recovered Signal threads set to erase themselves, and within a week all three learned they were out.
Korbak rejects the implication. Vanishing messages, he contends, are standard practice across Silicon Valley, and a tool used that broadly cannot reasonably be treated as covert.
What the three believe united their cases was lobbying for models that outsiders can inspect. Accept that premise and the precedent gets uncomfortable, because outside auditors only function when employees feel free to speak.
OpenAI rejects the link. The result is a staffing dispute carrying public weight at a moment when regulators in multiple capitals want lab safety work opened to independent review.
