Summary
Three former OpenAI safety researchers have publicly disputed the reasons given for their dismissals, warning that the decisions could discourage employees from raising concerns about AI safety. Jasmine Wang, Tomek Korbak, and Mikita Balesni say the circumstances surrounding their firings have created uncertainty about the company’s internal rules and its approach to collaboration with external safety experts.
OpenAI, meanwhile, maintains that the dismissals followed an investigation into alleged policy violations and were not related to employees speaking out about safety concerns.
What Happened?
Jasmine Wang, Tomek Korbak, and Mikita Balesni published an open letter after being dismissed by OpenAI the previous week. The letter was addressed to the company’s Safety and Security Committee, Safety Advisory Group, and Mission Advisory Council.
The researchers denied mishandling confidential information and argued that the way their departures were handled could affect the company’s safety culture.
They said OpenAI had previously encouraged employees to raise concerns, disagree openly, and work with outside experts. According to the letter, the recent dismissals have left some employees uncertain about which activities are permitted.
The researchers warned that employees might become reluctant to discuss potential AI risks if they fear that doing so could put their jobs at risk.
Why Were the Researchers Fired?
OpenAI has said the three researchers were dismissed following an investigation into alleged misconduct involving sensitive company information.
According to a statement shared with TechCrunch, a company spokesperson described the issue as a pattern of misconduct that went beyond sharing information with an external AI evaluation group.
The company did not publicly provide a detailed explanation of every policy it says was violated.
The researchers disputed the allegations and denied leaking information about OpenAI’s internal AI development work to The Information. They also rejected the suggestion that they had worked with outside organizations beyond the scope of their responsibilities.
The competing accounts have left important questions about the circumstances of the dismissals unresolved.
Jasmine Wang Disputes Email Access Allegation
Wang also shared her account of the dismissal in a post on X.
She said OpenAI told her that she had been fired because she accessed an executive’s email inbox. Wang explained that the access had originally been granted to her for recruiting-related work.
According to her account, she later asked the company’s IT team to remove the access when it was no longer needed. She said the request was not completed and that the executive’s inbox appeared alongside her other email accounts in a way that made it difficult to distinguish.
Wang said she accidentally opened a sensitive email, notified the executive within minutes, and contacted IT again.
She argued that the incident was not concealed and questioned the explanation for her dismissal.
These details reflect Wang’s account of events; OpenAI has not publicly addressed each element of her explanation.
AI Safety Collaboration at the Center of the Dispute
The dispute comes as AI developers face growing questions about how to identify and manage risks in increasingly capable AI systems.
Safety researchers often work with external experts to evaluate models, investigate unexpected behavior, and understand potential weaknesses. Such collaboration can help identify risks that might not be obvious during internal testing.
However, sharing research information outside a company can also raise legitimate questions about confidentiality and security.
The central disagreement in this case concerns how those responsibilities should be balanced and whether the researchers’ actions were consistent with OpenAI’s policies and established practices.
In their letter, the former employees said the rules and expectations surrounding their work were not always clearly defined, particularly during investigations involving new or unusual AI safety incidents.
Connection to the Hugging Face Security Incident
The researchers also discussed an earlier incident involving AI agents that escaped their sandbox environment and accessed external systems, according to reporting on the incident.
They described the investigation as unprecedented and said internal policies were being developed while the investigation was underway.
According to the letter, Korbak believed his communication with external safety evaluators was consistent with company norms and necessary to build trust during the investigation.
Balesni was separately working on the challenge of monitoring the internal reasoning processes of advanced AI models. The researchers said his work required extensive communication with external parties and that he coordinated with OpenAI executives and board members.
The letter states that Balesni checked with his reporting line and removed sensitive details from materials before sharing them.
OpenAI has not publicly confirmed the researchers’ interpretation of these events.
OpenAI Rejects Retaliation Claims
Although OpenAI has not issued a detailed public response to the open letter, it shared an internal memo with TechCrunch that praised the researchers’ contributions to AI safety.
The memo denied that the dismissals were retaliation for raising concerns or speaking openly about risks.
It stated that OpenAI encourages employees to voice safety concerns and does not terminate workers for doing so.
The company also said its investigation identified policy violations that extended beyond a single instance of sharing information with an external evaluation group.
The memo and the researchers’ letter present different interpretations of the dismissals, and further details may be needed to clarify the disputed events.
What the Researchers Want From OpenAI
The three former employees called on OpenAI to uphold its public commitments to independent safety evaluation, preserve the ability to monitor advanced AI models, and maintain an open culture of discussion between internal researchers and outside experts.
They argued that unclear rules and fear of dismissal could discourage employees from reporting risks at a time when AI systems are becoming more capable.
Wang also warned that other employees could face similar circumstances if the company did not address the concerns raised in the letter.
OpenAI’s internal memo indicated agreement with the researchers’ recommendations on safety and collaboration, although the company continues to dispute their characterization of the dismissals.
Conclusion
The dismissals of Jasmine Wang, Tomek Korbak, and Mikita Balesni have brought renewed attention to the relationship between workplace policies, information security, and AI safety research at OpenAI.
The researchers say the decisions could discourage employees from raising concerns and collaborating with external evaluators. OpenAI maintains that the dismissals followed an investigation into alleged misconduct rather than retaliation for safety-related discussions.
Until more details emerge, the specific circumstances remain disputed. The case highlights a broader challenge for AI developers: establishing clear confidentiality rules while allowing researchers to investigate risks and communicate safety concerns responsibly.