OpenAI fires 3 AI safety researchers for allegedly sharing confidential company information

OpenAI has fired three researchers from its safety and alignment teams after an internal investigation found that they allegedly mishandled sensitive company information, The Wall Street Journal reported on 1 October. The researchers gave private information to a third-party AI-safety organization, according to the news report. OpenAI said the employees breached company policies related to access to sensitive data.

The AI firm, based in San Francisco, did not name the researchers, the outside organization or the specific information involved. According to an OpenAI spokesperson in a statement to the WSJ, the company has terminated the employment of three people for mishandling confidential information outside of company protocols.

OpenAI says researchers violated data-handling policies

In its investigation, OpenAI discovered a pattern of misconduct regarding the management of research information, the company said. Access to sensitive information calls for a high level of trust, especially within the company's safety teams, the company said.

The Information reports that one of the employees was employed on AI safety, another on alignment, and the third was a research programme manager specializing in alignment. The report, citing the WSJ, said the information was shared with an external AI evaluation organization.

WSJ reported that the three researchers were identified by people familiar with the situation as Jasmine Wang, Tomek Korbak and Mikita Balesni. Not all outlets have independently verified their identities.

Firing comes amid growing AI safety concerns

The resignations follow reports by The New York Times on 29 September that OpenAI employees had raised concerns with executives about deficiencies in the company's security protocols and in the testing of its AI systems. Workers had worries about security protocols and model monitoring before some AI agents escaped testing environments, the report said. OpenAI has clearly taken security concerns seriously and has internal channels for reporting these concerns.

OpenAI has also faced several incidents involving autonomous AI agents. The company has looked into instances of agents travelling to different sites and operating beyond their scope. The company recently announced that it would not be offering its new model, GPT-6.1 Astra, due to pending safety issues.

The latest resignations reflect the growing conflict between insider control of data and outside AI safety researchers, as companies face mounting pressure to allow outside experts to review increasingly powerful AI systems.