OpenAI Fires Three Researchers
OpenAI fires researchers over a sensitive-information probe; researchers warned the move could chill safety work and raise model-monitoring risk.
KEY TAKEAWAYS
- OpenAI dismissed three safety researchers after an internal probe citing sensitive-information policy violations.
- The researchers denied misconduct and warned the move could chill internal safety work and external collaboration.
- They urged preserving model monitoring and independent audits to oversee advanced models' reasoning processes.
HIGH POTENTIAL TRADES SENT DIRECTLY TO YOUR INBOX
Add your email to receive our free daily newsletter. No spam, unsubscribe anytime.
OpenAI fired three safety researchers the week before October 9, 2026, after an internal investigation found violations of its sensitive-information policy. The researchers denied misconduct and warned the dismissals could chill safety research and external collaboration.
Company Account and Internal Memo
OpenAI dismissed Jasmine Wang, Tomek Korbak, and Mikita Balesni, all of whom worked on safety and alignment teams. The company said the researchers mishandled sensitive information outside established procedures, breaching company policy and trust.
An internal research-leadership memo stated the dismissals were unrelated to raising safety concerns or speaking out. It emphasized that employees are not terminated for raising such issues.
Researchers’ Response and Safety Concerns
On October 8, the three researchers published an open letter to OpenAI’s Safety and Security Committee, Safety Advisory Group, and Mission Advisory Council disputing the company’s account. They warned the firings could discourage internal debate, safety research, and collaboration with outside experts.
Jasmine Wang said she was told her dismissal resulted from accessing an executive’s email account, which she said had been authorized for recruiting purposes. She said she requested removal of the access, accidentally opened a sensitive message, and reported it.
Reporting described the investigation as involving alleged sharing of confidential information with an outside AI-safety or evaluation organization. OpenAI said the misconduct extended beyond that issue. The researchers denied involvement in any leak about less-monitorable model architectures or unauthorized contact with external parties.
The former employees urged OpenAI to preserve the ability to monitor advanced models’ reasoning processes and to work with independent safety auditors. Reporting linked the dispute to concerns about model monitoring and recording a model’s reasoning process, often called the chain of thought.
Their challenge highlights a key operational issue for OpenAI: maintaining robust internal and third-party oversight as models grow more sophisticated while enforcing controls on sensitive information and access.
