OpenAI safety researchers dispute reasons for dismissals

OpenAI’s dismissal of three safety researchers has prompted a public dispute over whether the decision stemmed from alleged mishandling of sensitive information or was connected to their safety work and outside collaborations. The company says the firings were for policy violations, while the researchers say their safety concerns and work with external groups were implicated. The supplied accounts do not independently establish which explanation is correct. The former researchers, Mikita Balesni, Tomek Korbak and Jasmine Wang, published an open letter to OpenAI leadership and safety committees describing concerns about the dismissals and the effect they say they are having on colleagues. They called for continued collaboration with outside safety evaluators, the preservation of ways to monitor advanced models, and an open culture in which employees can raise concerns. The letter warns that reduced ability to monitor how models reason could weaken safety work. The researchers argued that cooperation with external experts is important as AI systems become harder to assess.