OpenAI Doubles Down on Firing Three Safety Researchers: 'Significant Breach of Trust,' But No Details
A company whose core narrative is "safety" fired three people who worked on safety — then said the details can't be shared because they involve sensitive information.
Three key facts
On October 23, OpenAI posted on X confirming it had fired three safety researchers: Jasmine Wang, Tomek Korbak, and Mikita Balesni. The company said an internal investigation found "a significant breach of trust" and violations of "clear policies on handling sensitive information." OpenAI specifically stressed the decision was not about the three raising concerns about the company and AI safety.
It was a response to the trio's open letter. According to reports, the researchers had published an open letter and posted on social media saying they believed they were fired for raising safety concerns, and that they had "acted in line with OpenAI's mission and within the working norms of the time." OpenAI said its investigation uncovered breaches "beyond what's outlined in the letter" — but provided no details.
The backdrop is rising internal safety anxiety across the industry. The report notes that after a series of high-profile security incidents this year — most notably OpenAI's Hugging Face hack — staff concerns about frontier-system safety have grown markedly, with many employees calling on companies to slow down and strengthen safeguards.
Wang Dou's Take
Let's be clear: nobody has the whole truth here right now. The company says "violated sensitive-information policy"; the employees say "because we raised safety concerns"; neither side has produced evidence a third party can verify. Either conclusion today is just one party's version.
But one detail deserves attention: that line about "breaches beyond what's in the letter, but we won't give details." In PR terms it does something very precise — it denies the most damaging charge ("we fired them for raising safety concerns") while shifting the burden of proof back onto a party that can't defend itself. You cannot rebut an accusation whose content you don't know. That is the near-universal structure of every "internal investigation" statement.
What the whole industry should really watch is the timing. Within one year: multiple AI security incidents, third-party red teams finding monitoring holes, employees calling for a slowdown — and precisely at this juncture, the people leaving are the ones who did safety. Whatever the specific reason, the externally visible fact is this: the people saying "we should slow down" are having their seats cleared.
One last line for everyone doing safety work at an AI company: if your job is "pointing out problems" and your employer is "the problem being pointed out," what you need may not be better communication skills, but a tougher employment contract and a communication channel that leaves a record. Disappearing messages disappear automatically. This story won't.
Source: The Verge · OpenAI Statement