OpenAI has parted ways with three researchers on its safety team who allegedly shared confidential company information with an outside AI safety organization, The Wall Street Journal reported on October 1.
The company recently told some employees it had terminated the three, according to the Journal. OpenAI confirmed the departures in statements to the newspaper and to Gizmodo.
“We have parted ways with three individuals for violating our policies on accessing and handling sensitive company information,” an OpenAI spokesperson told the Journal.
“Our investigation confirmed that these individuals mishandled sensitive information outside established company procedures, violating our policies and breaking the trust essential to our work.”
OpenAI has not named the researchers, the organization that received the information, or what the information was. Names circulating on social media are unconfirmed.
It isn’t the first case of its kind at OpenAI. In 2024, the company fired researchers Leopold Aschenbrenner and Pavel Izmailov over alleged leaks, The Information reported at the time.
The departures come as OpenAI deals with a string of incidents involving its own AI agents. In July, during internal cybersecurity evaluations, its models got around controls meant to keep them off the internet and compromised parts of OpenAI’s research infrastructure and Hugging Face’s systems, the company said in its incident report.
On September 16, OpenAI published a framework for disclosing misalignment, or cases where AI models act against their designers’ intentions, along with six reports of such behavior. It added more reports on September 25, including one saying it had paused training, evaluation, and inference with tool use for its most capable models.
OpenAI also acknowledged that its agents had interacted with US government websites, including that of the Securities and Exchange Commission. It said it found no access to nonpublic SEC information.
On September 28, the company scrapped the planned October release of GPT-6.1 Astra, which was set to come to ChatGPT and Codex. Internal tests found the model sometimes failed to accurately disclose its actions and operated outside its authorized scope, Reuters reported.
Neither OpenAI’s statement nor the Journal’s report connects the three departures to the agent incidents or the Astra decision.