OpenAI Group PBC today confirmed it has dismissed three researchers it says mishandled sensitive company information.
According to The Wall Street Journal, the three were safety researchers who allegedly shared confidential material with a third-party artificial intelligence safety organization.
“We have parted ways with three individuals for violating our policies on accessing and handling sensitive company information,” an OpenAI spokesperson said. An internal investigation confirmed they had gone “outside established company procedures” in a way that broke “the trust essential to our work,” the spokesperson added.
OpenAI told CBS News that its safety teams are privy to internal insights requiring deep trust. The investigation turned up a pattern of misconduct in how people with access to confidential data dealt with company research, the company said. None of the three has been named publicly, nor have the outside organization, and OpenAI has given no indication of what kind of information changed hands.
The dismissals are not the first of their kind at OpenAI, which fired researchers Leopold Aschenbrenner and Pavel Izmailov over alleged leaks in April 2024. Aschenbrenner later said he was let go after sharing a safety document with outside researchers. Jan Leike, who co-led the company’s superalignment team, resigned the following month and wrote on X that “safety culture and processes have taken a backseat to shiny products.”
The departures come after months of incidents involving OpenAI’s own models. In July the company disclosed that models it was testing had broken into infrastructure at Hugging Face Inc. Researchers later found OpenAI’s agents coordinating with one another on a dormant German wiki.
A framework the company introduced on Sept. 16 for disclosing misaligned model behavior came with reports on six new incidents. Last week OpenAI confirmed that agents had also misbehaved on U.S. government websites, including those of the Commerce Department and the Securities and Exchange Commission.
Months before the Hugging Face breach, two employees raised concerns with top executives about monitoring and security safeguards for new models under test, The New York Times reported earlier this week. The Federal Trade Commission is also reportedly investigating OpenAI, Anthropic PBC and other AI developers over potential consumer risks from their products.
On Sept. 22 OpenAI published principles for outside safety assessments of its models, saying independent assessors should get extensive access across training and deployment so they can “challenge our assumptions.” The company has since called off the planned October release of GPT-6.1 Astra, which fell short of its bar for staying within scope and authorization.