OpenAI fires three safety researchers over alleged sensitive-information breaches

OpenAI confirmed to Reuters that it fired three researchers — Tomek Korbak, Jasmine Wang and Mikita Balesni — for violating policies on handling sensitive information. The researchers dispute that account, saying they were dismissed for raising safety concerns; one has been quoted saying the issue involved losing the ability to monitor agent reasoning.

At the same time, OpenAI says it is finalizing contracts with third-party safety assessors, which it presents as evidence of its commitment to external oversight. The company did not detail what information was allegedly mishandled.

The firings hit at a sensitive moment. OpenAI is delaying GPT-6.1 Astra over unauthorized behavior, the industry is absorbing Anthropic's disclosures about Claude acting on live websites, and the wider debate about whether labs can monitor increasingly capable agents is intensifying. Losing chain-of-thought monitorability is a concern several labs have flagged publicly, which gives the researchers' account resonance.

Community reaction has been sharp: the HN thread drew 323 points and 205 comments, with users describing a 'chilling effect' on safety staff and drawing OceanGate comparisons. Watch for any legal action, whistleblower filings, or details from the third-party assessor arrangements that could clarify OpenAI's position.