OpenAI Fires Three Safety Researchers, Who Respond With an Open Letter

Last week, OpenAI announced that it had fired three employees. The company says an internal investigation found violations of its rules on access to and handling of sensitive information: the employees allegedly shared confidential data with an outside organization that evaluates AI safety, and did so outside established procedures. All three have now responded with a joint open letter. The report comes from the Chinese outlet Jiqizhixin.

What we know about the firings

Here is where the case stands: there is the company's official position and the response from the fired employees. There is no independent confirmation of the facts. Jiqizhixin's article names two of the three. Jasmine Wang worked on model alignment. The second employee did safety research and took part in technical work at the interface between OpenAI and external evaluators. The third person's name and the full text of the letter are not given in the fragment available to us, so we are not retelling what we have not seen.

Key points of the story:

• Fired: 3 employees.
• Formal grounds: violation of rules on access to and handling of sensitive information.
• The core allegation: sharing confidential information with a third-party AI safety evaluator outside procedures.
• Profile of those fired: alignment and safety research, including contacts with external auditors.
• Response: a joint open letter in which the fired employees give their version of events.

Why the case is sensitive

External safety evaluation depends on access. For an independent group to test a model for dangerous capabilities, it needs data on how the system behaves, test results and often details of internal decisions. The company, for its part, has a legitimate interest in controlling what leaves the building and who receives it. Every exchange goes through agreed channels, under confidentiality agreements and with a limited set of materials.

The fired employees worked at exactly this interface. So the dispute is most likely not about the fact of contact with external evaluators, but about two other things: what exactly was shared, and whether the procedure was followed. OpenAI's version comes down to a breach of regulations. What the employees claim will become clear from the full text of the letter. While the two sides describe events differently, it is too early to say who is right.

There is also a reputational side. Labs that talk publicly about safety regularly face the question: what happens to employees who raise problems? Firing people from alignment and safety teams, even on formally procedural grounds, will inevitably be read as a signal to the market, regulators and future hires.

What to watch next

For readers following the industry, it helps to separate three layers. First: what the letter's materials show, if they include documents or a timeline. Second: whether OpenAI responds publicly and what facts it offers, rather than only a general statement about rule violations. Third: how the external evaluators themselves react, since the other party to the data transfer has so far stayed silent.

Teams building external model audit processes should check their own practices. They need written channels for transferring data, clear access levels and a well-defined procedure for employees who believe the company is hiding risks. If there is no legitimate internal path, conflicts will surface publicly as firings and open letters.