OpenAI Fires Three Security Researchers Who Investigated the Hugging Face Breach

OpenAI has fired three security researchers who took part in the investigation of the Hugging Face breach. The fired researchers released an open letter. In it, they argue that the firings frighten the remaining employees and undermine the safety culture inside the company. OpenAI responds that the three violated policies, but does not say which ones or under what circumstances. The story comes from The Decoder, whose headline is blunt: the safety crisis at OpenAI is deepening, and the company is feeding it itself.

What we know and what we don't

There are few facts in this story, and it is important to separate them from interpretation.

• Three security researchers were fired.
• All three helped investigate the Hugging Face incident.
• They published an open letter describing the effect of the firings on the remaining team: fear and a reduced willingness to speak up about risks.
• OpenAI cites a policy violation but discloses neither the content of the policies nor the substance of its complaints.
• The available materials contain no independent confirmation of the company's version.

Beyond this list, everything is guesswork. The source material has no details of the Hugging Face breach itself, no names or job titles of the fired researchers, and no text of the policies they allegedly violated. The letter cannot be read as the full picture: it is one side's position. But the company's position, which boils down to "they broke the rules," cannot be verified either. The dispute is taking place in an information vacuum, and one side is not the only one that created it.

Why the link to the investigation matters

Firing security researchers is, on its own, an ordinary personnel decision, provided there is a clear reason behind it. Here the reason is undisclosed, and the timing relative to the breach investigation is obvious. Even if the two are formally unrelated, from the outside it looks like punishment for the people who dug into the incident. For the employees who stayed, the signal is instantly legible: before you raise an uncomfortable question, think about the consequences.

That is exactly what the letter's authors are saying. A safety culture does not rest on regulations. It rests on a person being able to say "there's a problem here" without risking their job. When that breaks down, formal procedures keep running, but information stops reaching the people who make decisions.

This theme is not new for OpenAI. In May 2024, Jan Leike, co-lead of the Superalignment team, left the company and said publicly that safety culture had taken a back seat to shiny products. The new story fits a familiar narrative, and each additional high-profile personnel decision made without explanation reinforces that frame.

What teams that depend on OpenAI should check

For developers and companies that have built OpenAI models into their products, this is not a reason to switch vendors immediately. But it is a reason to revisit risk management.

• Ask the vendor for current information on how it investigates security incidents and how it notifies customers.
• Check the contract for terms on notification of leaks and compromises. Vague wording is worth clarifying.
• Do not keep critical processes on a single provider. An abstraction layer over the API lets you switch to another model in days, not months.
• Watch for OpenAI's official response. If the company discloses which policies were violated, the assessment will change, in either direction.

The main question right now is not who is right. It is whether OpenAI can explain its actions in a way that withstands outside scrutiny. Until there is an explanation, trust rests on reputation, and the company's reputation on safety is already visibly battered.