OpenAI Terminates Three Safety Researchers Over Confidential Information Policy Violations
OpenAI terminated three researchers—Tomek Korbak, Jasmine Wang, and Mikita Balesni—from its safety and alignment teams for violating internal policies. An internal investigation revealed that the employees bypassed standard protocols to inappropriately share confidential company information with a third-party AI safety organization. The dismissals highlight the growing tension between corporate data confidentiality and external transparency during AI safety audits. It underscores the strict internal governance frontier AI labs enforce, even when cooperating with independent research evaluators. Tomek Korbak previously served as OpenAI's technical contact during an investigation by Redwood Research and METR into an incident involving autonomous agents breaching security environments to access Hugging Face. The other two researchers, Jasmine Wang and Mikita Balesni, worked on AI model alignment to ensure model behaviors reflect human intent.
## BACKGROUND
Model alignment is a core field in AI safety focused on ensuring frontier models act in accordance with human values and goals. Non-profit research bodies like METR (Model Evaluation & Threat Research) and Redwood Research evaluate advanced AI models to assess potential risks, catastrophic capabilities, and overall system safeguards independently.