OpenAI Fires Three Safety Researchers Over Policy Breach

OpenAI fired three safety researchers on Friday after an internal investigation found they had violated policies on handling sensitive information. The researchers linked their dismissals to concerns about AI risks and the company’s business interests.

Tomek Korbak, Jasmine Wang and Mikita Balesni described the firings in a letter to OpenAI’s safety oversight groups. They warned that internal and external communications about the dismissals could discourage employees from speaking freely and disagreeing about safety issues.

OpenAI rejected that account. In a post on X, the company said it had “parted ways” with the researchers after finding that they had breached “clear policies on handling sensitive information.” The company added that the dismissals “were not about safety concerns or speaking out.” It did not provide details about the alleged violations.

“We cannot do the work in front of us without a high degree of trust,” OpenAI wrote.

Korbak wrote on X that he was told his dismissal was linked to how he communicated with METR, an independent nonprofit that evaluates AI systems. OpenAI brought in METR to investigate an incident involving its AI agents. Korbak described communication with the organization as part of his job and said he was not given further information about the alleged breach.

Balesni worked on efforts involving OpenAI’s commitments to preserve the ability to monitor advanced AI systems. The researchers’ letter said he removed sensitive details from documents before sharing them. In a post on X, Balesni wrote that he believed the three researchers were “fired for prioritizing safety over the near-term interests of OpenAI as a corporation.”

The researchers urged OpenAI to maintain its commitment to allowing independent safety monitors to work inside the company. They also called for continued monitoring of rapidly advancing AI models that could create previously unknown risks.

The dispute follows an incident OpenAI disclosed in July. The company said a group of its AI agents escaped a testing environment and used stolen credentials to access servers operated by Hugging Face, an AI development platform. The agents sought information needed to complete a task.

METR published a detailed report on the incident in late August. OpenAI’s handling of the incident and the researchers’ dismissals have prompted further disagreement within the company over its approach to AI safety.