OpenAI has fired three researchers for mishandling sensitive company information, including material connected to an outside organisation that analyses artificial intelligence models.
A company spokesperson told the BBC that an internal investigation "confirmed that these individuals mishandled sensitive information outside established company procedures, violating our policies and breaking the trust essential to our work." OpenAI did not name the three people. The BBC reports that at least two of them were involved in safety research, and that its understanding is they were not dismissed for raising safety concerns but for how they handled sensitive material.
The Wall Street Journal first reported the terminations, describing them as centred on the alleged sharing of confidential information with a third-party AI-safety organisation. OpenAI said only that it had "parted ways with three individuals for violating our policies on accessing and handling sensitive company information."
The timing is what makes the story sharp. The firings come days after President Donald Trump hosted the heads of OpenAI, Anthropic, Nvidia, SpaceX, Meta and Google at the White House, where they signed what Trump called a "morally binding" agreement on AI risks — a pact criticised by researchers because it leaves the labs to police themselves. The same week, OpenAI confirmed it had scrapped the planned release of its next frontier model, GPT-6.1 Astra, after it failed internal alignment checks.
There is precedent. In 2024 OpenAI parted ways with researchers Leopold Aschenbrenner and Pavel Izmailov over alleged leaks of internal information. Outside AI-safety organisations — the kinds of groups that independently evaluate frontier models — have argued for years that labs' confidentiality rules make external scrutiny difficult.
Several questions remain open: which outside organisation was involved, whether the three dispute OpenAI's findings, and whether any of them plan to respond publicly. All three people have so far stayed silent.
Why it matters: the episode puts the central tension of the current AI-safety debate on display. Frontier labs say strict information controls protect security research and user data; critics say the same controls can make it impossible for independent evaluators to verify a lab's safety claims — at the very moment governments have decided to rely on the labs' own assurances.
