Three former OpenAI safety researchers, Jasmine Wang, Tomek Korbak and Mikita Balesni, have published an open letter calling for clearer internal procedures around AI safety research and collaboration with independent experts.
The researchers said their work focused on identifying emerging risks in advanced AI systems, including questions around model monitorability, external evaluation and responsible information-sharing practices. They argued that well-defined processes are essential for teams working on fast-evolving technical challenges.
Focus on transparent safety practices
In their letter, the researchers disputed claims that they had handled sensitive information improperly. They said their interactions with external safety specialists were intended to support responsible research and strengthen accountability across the wider AI ecosystem.
OpenAI has stated that its employment decisions were not related to employees raising safety concerns. An internal message attributed to company leadership also recognized the researchers' contributions to AI safety and reaffirmed the organization's support for open discussion of technical risks.
The discussion has drawn attention to a broader industry challenge: how frontier AI developers can balance data protection, internal security and meaningful cooperation with outside evaluators. As models become more capable, researchers increasingly emphasize the value of shared standards, auditable processes and channels for constructive technical feedback.
A growing role for independent evaluation
The former researchers encouraged OpenAI to continue developing mechanisms for third-party safety assessment, preserve the ability to evaluate advanced systems and maintain dialogue between internal teams and the global research community.
Clearer governance frameworks could help AI organizations align innovation with public trust, ensuring that safety expertise remains central as increasingly capable systems move from research environments into everyday life.
