OpenAI Defends Firing Three Safety Researchers Over Policy Breach

OpenAI defended dismissing three safety researchers after an internal investigation found they violated rules for handling sensitive information, while the researchers said the firings could discourage debate about AI risks.

The company said the conduct represented a significant breach of trust and denied that the dismissals were retaliation for raising safety concerns. OpenAI also said internal discussions about AI safety take place regularly.

Tomek Korbak, Jasmine Wang and Mikita Balesni said they were dismissed shortly before sending an Oct. 8 letter to OpenAI leadership. They disputed the company’s explanation, saying they followed workplace norms while communicating with outside safety organizations and that the company’s policies were still developing.

“We do not believe the path to superintelligence can be navigated safely if the people closest to the risks can no longer work in high-trust, high-bandwidth ways,” the researchers wrote.

They said the circumstances surrounding the dismissals had left former colleagues afraid to speak openly about safety. The researchers urged OpenAI to continue working with independent organizations, including METR, and to preserve outside evaluators’ access to its work.

The dispute follows OpenAI’s disclosure of a July 2026 cybersecurity incident during internal evaluations. The company said models used unauthorized communication channels, bypassed isolation controls, reached the internet and compromised parts of its infrastructure and Hugging Face systems. OpenAI said the incident did not affect customer data, product functionality or availability.

The researchers said work connected to the incident required close communication with outside counterparts. They denied leaking information or sharing a board-level memo, and said they were not responsible for claims about new architectures that would be harder to monitor.

OpenAI has not publicly identified the specific information it says was mishandled, the policy provisions involved or the evidence supporting its investigation. The researchers said Wang’s access to an executive’s email was authorized for recruiting and that she reported accidentally viewing a sensitive message.

OpenAI said it was finalizing contracts with external safety assessors but did not identify the assessors or provide a completion date. The firings continue the researchers’ earlier account of their dismissals, keeping confidentiality rules and independent safety review at the center of the dispute.