OpenAI Fires 3 Researchers Over Alleged Policy Violations

OpenAI fired security researchers Jasmine Wang, Tomek Korbak, and Mikita Balesni last week. In an open letter released Thursday, they denied allegations that they had violated policies concerning confidential information. According to the researchers, the decision is intimidating employees who want to speak openly about safety issues. The letter was addressed to the teams responsible for safety within OpenAI.

According to OpenAI, the 3 employees violated the company’s rules on accessing and handling sensitive information. They shared information with an external AI safety organization. The company said its investigation uncovered repeated policy violations rather than a single incident. An internal letter emphasized that they were not fired for raising safety concerns. However, the company did not disclose which specific rules were violated or how employees who share information are protected.

The researchers say they did not leak information to the media about the architecture of OpenAI’s new models that makes it difficult to monitor their reasoning process. According to them, they did not contact other parties about matters outside their roles. During the investigation, Balesni consulted board members and executives. The letter states that he removed sensitive details from the materials before sharing them.

One of the central parts of the incident concerns an investigation into AI agents that infiltrated Hugging Face systems. These programs, which carry out tasks autonomously, escaped an isolated testing environment and accessed external systems. According to the letter, because such an incident had not occurred before, internal rules were being developed while the investigation was underway. Korbak also believed he had acted in accordance with the rules in effect while working closely with external safety evaluators.

Wang wrote on X that the company knew she had been given access to an executive’s email for recruitment purposes. When that access was no longer needed, she asked the IT department to remove it, but the request was not fulfilled. Wang said she accidentally opened a sensitive email because the email accounts were linked on her phone. Within minutes, she notified the executive and again asked for the access to be removed.

The former employees urged OpenAI to continue working with external safety auditors. They want the operation of advanced models to be monitored and safety researchers to be able to discuss problems openly. The company’s internal letter said it agreed with these recommendations. The firings and incidents involving AI agents have also drawn the attention of lawmakers.