OpenAI said it fired three safety researchers and described the action as a response to a “breach of trust” and policy violations involving sensitive information, according to AP News and The Washington Post. The researchers — Tomek Korbak, Jasmine Wang, and Mikita Balesni — said in a public letter that they were fired “last week,” addressed their letter to OpenAI’s Safety and Security Committee, Safety Advisory Group, and Mission Advisory Council, and argued that the dismissals were chilling OpenAI’s open culture.
In that letter, the researchers said safety work depends on high-trust collaboration with external experts. They also denied being the source of the leak referenced by OpenAI and said they do not know who was. The letter further said Jasmine Wang’s access to an executive’s email had been delegated for recruiting purposes and that she asked for it to be removed when it was no longer needed.
The dispute is therefore not just about who handled information inside one company. It is also about what kinds of access, oversight, and outside collaboration a frontier AI lab will allow while it polices confidential material. The supplied sources do not identify the full contents of the underlying leak, the specific sensitive-information rules at issue, or any additional personnel actions beyond the three firings reported here.
What OpenAI said
AP News reported that OpenAI said it fired the three researchers. The Washington Post reported that OpenAI said the firings were not retaliation and that an internal investigation found a significant breach of trust. The research bundle also says OpenAI described the action as a response to a “breach of trust” and policy violations involving sensitive information.
Those statements are the company’s account in the supplied material. They establish that OpenAI tied the firings to information-handling concerns, but the bundle does not identify the evidence OpenAI relied on, the precise policy provisions it invoked, or any company explanation beyond those descriptions.
What the researchers said
The researchers’ letter presents a different account. It says the three were safety and alignment employees fired “last week,” and it asks OpenAI’s Safety and Security Committee, Safety Advisory Group, and Mission Advisory Council to intervene in the way the company handles safety work. In the letter, the researchers said the firings were chilling OpenAI’s open culture and that safety work depends on high-trust collaboration with external experts.
The letter also says they were not the source of the leak under dispute and that they do not know who was. It says Jasmine Wang’s email access had been delegated for recruiting purposes and that she asked for it to be removed when it was no longer needed. The supplied sources do not identify what the email access was used for beyond that recruiting-related delegation, or whether OpenAI accepted that explanation.
What the researchers asked for
The letter does not limit itself to protest. It asks OpenAI to embed third-party safety auditors, preserve monitorability of frontier models, and continue supporting open and transparent dialogue with the safety ecosystem. Those are requests in the supplied evidence; the bundle does not say whether OpenAI agreed, rejected them, or responded to them in the sources provided.
Because the bundle does not state that any of those requests were granted, the careful reading is simply that the former employees sought stronger external oversight and continuing transparency. The sources supplied here do not identify any timetable for those requests, any formal process for evaluating them, or any decision by OpenAI on them.
Why this matters for AI labor and governance
The immediate labor effect identified in the research bundle is the loss of three safety-and-alignment roles at OpenAI. Beyond that direct job loss, the dispute is a governance test for a company that says it wants to manage sensitive information while maintaining safety research. The AP News report says the researchers accused the company of putting corporate interests before safety, while the letter itself says open safety work relies on external collaboration.
The bundle’s labor-impact note says this episode may make AI safety jobs less attractive to workers who expect broad external collaboration, while making employers more likely to tighten access, review, and disclosure rules for researchers. That is analysis included in the research bundle, not a verified statement from OpenAI or the researchers. The supplied sources do not identify broader hiring plans, any workforce-composition change at OpenAI beyond the three firings, or any company-wide rule changes following the dispute.
What to watch next
The next relevant developments, based on the supplied material, are limited and specific: whether OpenAI provides any further explanation of the “breach of trust” and policy-violation claims; whether the researchers or their supporters press the request for third-party auditors and continued model monitorability; and whether any additional public statements clarify the scope of the alleged leak or the role of delegated email access. The sources supplied here do not identify a hearing, arbitration, audit, or other proceeding already underway.
For now, the record in the bundle shows a direct clash between a company’s control over sensitive information and former employees’ claim that safety work requires openness, outside review, and high-trust collaboration. The dispute is significant because it sits at the intersection of employment discipline and AI governance, but the supplied sources do not settle the factual disagreement between the two sides.
Sources
- OpenAI fires 3 safety researchers in dispute over AI risks — AP News
- OpenAI fires three safety researchers, cites information policy breach — The Washington Post
- OpenAI cannot make AI safe on its own — Mikita Balesni