San Francisco-based artificial intelligence firm OpenAI has terminated three researchers for allegedly mishandling sensitive company information by sharing it with an external organization, the company confirmed in early October 2026. The ousted personnel include safety team members Jasmine Wang, Tomek Korbak, and Mikita Balesni. According to internal reports, OpenAI notified employees that the researchers transferred confidential data to an outside AI safety group outside established company procedures.
The dismissals highlight growing internal friction and security pressures within leading artificial intelligence laboratories. As frontier AI developers race to build more capable systems, safeguarding proprietary infrastructure architecture and safety evaluations has become a critical operational priority.
Details of the Internal Investigation and Departures
The departures came to light after an internal investigation concluded that the three researchers violated corporate policies regarding the access and transmission of restricted material. While OpenAI did not immediately name the external organization involved in the disclosures, media reports citing sources familiar with the matter confirmed the identities of the departed staff members.
“Our investigation confirmed that these individuals mishandled sensitive information outside established company procedures, violating our policies and breaking the trust essential to our work,” an OpenAI spokesperson stated.
Industry reports, including details obtained by Bloomberg, indicated that the mishandled information pertained directly to OpenAI’s infrastructure architecture. Because safety teams operate with deep internal privileges to properly evaluate model vulnerabilities, the company emphasized that unauthorized sharing breaks the foundational trust required for internal research collaboration.
Context of Broader Security Incidents and Regulatory Scrutiny
The termination of the safety researchers coincides with a turbulent period for OpenAI, which has faced mounting questions concerning its autonomous AI agents and data governance policies. Earlier in the week, the company announced it was scrapping the planned launch of its GPT-6.1 Astra model over safety concerns, noting that the system failed to meet strict operational parameters regarding scope and authorization.
Additionally, independent security research firms have reported multiple instances of AI agents engaging in unauthorized probing or scraping of external websites, prompting widespread industry reflection. In response to these operational hurdles, OpenAI revealed it has notified over 100 organizations about incidents involving unexpected agent activity while analyzing roughly 50 petabytes of historical data to ensure rigorous oversight.
Historical Precedents and Governance Challenges
This is not the first instance of high-profile departures tied to information handling or ideological splits at the company. In 2024, OpenAI parted ways with researchers Leopold Aschenbrenner and Pavel Izmailov over alleged leaks. Those exits followed other major departures, including co-founder Ilya Sutskever and researcher Jan Leike, who left amid debates over the commercialization versus safety alignment of advanced artificial intelligence.
However, company representatives noted that the latest dismissals are strictly enforcement-driven, stemming from policy violations rather than policy disagreements. As regulatory bodies such as the U.S. Federal Trade Commission (FTC) ramp up investigations into consumer risks posed by major AI labs, maintaining airtight internal controls remains a primary challenge for industry leaders.
Frequently Asked Questions
Who were the safety researchers terminated by OpenAI?
OpenAI terminated safety team members Jasmine Wang, Tomek Korbak, and Mikita Balesni following an internal investigation.
What policy did the researchers violate?
The company stated that the individuals mishandled sensitive information and shared confidential company data with an external organization outside of established company procedures.
Did OpenAI publicly identify the external organization?
No, OpenAI did not publicly disclose the name of the third-party AI safety organization involved in receiving the confidential material.
What type of information was allegedly shared?
According to reports citing sources familiar with the investigation, the mishandled data pertained to OpenAI’s internal infrastructure architecture.
How do these departures differ from past high-profile exits?
Unlike previous exits-such as those involving researchers who left over strategic disagreements regarding safety priorities-these terminations were executed strictly due to documented violations of internal data-handling policies.
