Two researchers fired by OpenAI in early October are publicly contradicting the company's characterization of their departures, marking an unusual moment of transparency in a tech industry often shrouded in confidentiality agreements and legal caution.
According to AI Weekly, Mikita Balesni and Tomek Korbak disputed OpenAI's public statement that their terminations resulted from a "significant breach of trust." In their first recorded remarks since the firings, the pair appeared on NPR's Machine Gods podcast and provided an alternative account of the events leading to their exits.
What Actually Happened
The researchers claimed the real source of tension centered on how safety personnel within OpenAI interact with external auditors and oversight bodies. Rather than any singular violation of policy or misconduct, the pair described a fundamental disagreement about the company's approach to transparency and external accountability in its safety practices.
Korbak, who held the position of technical lead for safety research at OpenAI, emphasized that the narrative presented publicly did not align with conversations he and Balesni had with company leadership during their termination.
Broader Questions About AI Governance
The dispute highlights ongoing tensions within major AI labs regarding how safety research should be conducted and shared. As AI systems grow more powerful, questions about internal safety standards and how companies communicate with external regulators, researchers, and auditors have become increasingly fraught.
- Internal safety research practices remain largely opaque to outside stakeholders
- Communication protocols between labs and external auditors lack clear industry standards
- Researchers face pressure to balance transparency with competitive and legal concerns
Implications for AI Industry Accountability
The incident underscores a recurring challenge in the artificial intelligence field: the absence of standardized frameworks for how companies should handle safety research and external oversight. When disputes arise, companies and former employees often operate under different interpretations of what constitutes breach, misconduct, or acceptable behavior.
The researchers' willingness to speak on record represents a rare break from the typical pattern where employment disputes between tech companies and their staff remain confidential. Their account suggests that OpenAI's safety division may be grappling with questions about how much internal research should be disclosed to external parties, a question with significant implications for AI governance and public trust.
Industry observers have long noted that major AI companies maintain tight control over safety research findings, citing competitive and security concerns. This case indicates those concerns may be creating friction within organizations between researchers committed to transparency and leadership focused on protecting proprietary information.
Whether this public disagreement prompts changes to how OpenAI manages external communications or handles safety researcher relationships remains unclear. The incident does, however, demonstrate that the conversation about AI safety and corporate accountability is increasingly difficult for companies to manage through press statements alone.



