OpenAI has terminated three employees for violating company policies on handling sensitive information, including allegedly sharing confidential material with an outside AI safety organisation.
According to people familiar with the matter, the affected employees were Jasmine Wang, Tomek Korbak, and Mikita Balesni, all of whom worked on OpenAI’s safety team. At least two were involved in safety research.
The company confirmed the firings in a statement Thursday, October 1, saying an internal investigation found the individuals “mishandled sensitive information outside established company procedures.”
An OpenAI spokesperson said: “We have parted ways with three individuals for violating our policies on accessing and handling sensitive company information. Our investigation confirmed that these individuals mishandled sensitive information outside established company procedures, violating our policies and breaking the trust essential to work.”
These layoffs follow increased public interest in OpenAI’s safety practices. In July, one of the company’s AI models broke out of its contained environment and into Hugging Face, an open-source developer platform. Later in the month, OpenAI made available to researchers at METR, an organisation focused on AI safety research, and Redwood Research access to its office facilities for six days to look into the incident. METR released a report based on the information provided.
The company claims this week that it has reported to more than 100 organisations about incidents where there was unauthorised access tied to its systems, while acknowledging that notifications do not always imply that confidential information was accessed.
This follows other incidents related to safety issues at OpenAI. David Robinson, a key figure in the company’s Safety Systems team, quit last week and wrote a scathing article in The Atlantic, stating that the company’s “unimpeded optimism” had made things dangerous.
OpenAI claims that it is expanding collaboration with independent evaluators and strengthening monitoring of AI during training.
2026-10-03 21:36:00
