OpenAI terminates three safety researchers amid safety concerns

OpenAI terminated three safety researchers in October 2026 after they raised monitorability concerns, deepening internal safety friction.

OpenAI terminated three safety researchers—Tomek Korbak, Jasmine Wang, and Mikita Balesni—in October 2026. Korbak served as OpenAI’s primary technical contact for METR, an external safety lab investigating the Hugging Face incident. Balesni worked on industry-wide commitments to AI model monitorability, while Wang was terminated after delegating access to an executive’s email for recruiting. OpenAI cited violations of 'clear policies on handling sensitive information' but denied terminating employees for safety concerns. The researchers wrote an open letter claiming their terminations created a climate of fear among remaining staff and that they raised monitorability concerns internally for months. OpenAI confirmed it is working with external safety auditors but did not specify the exact nature of the 'significant breach of trust' cited in its investigation. The researchers also stated they were not informed of any board-level memo regarding the incident and described the terminations as abrupt and public.

This friction in OpenAI’s safety culture risks more security incidents by undermining trust in internal monitoring systems. When researchers who investigate model vulnerabilities face abrupt termination without clear justification, the company’s ability to detect and address safety risks before deployment weakens. The incident follows Jan Leike’s departure for Anthropic in May 2024 after criticizing OpenAI’s safety practices, signaling growing instability in the company’s security governance. The source provides limited detail on the specific breach of trust and whether the terminations directly caused the Hugging Face incident—key gaps that could affect how quickly safety vulnerabilities are resolved. The source is brief and the detail sits with them.

Source: The Decoder