Fired OpenAI Safety Researchers Deny Leaking Secrets

Three OpenAI safety researchers fired for suspected leaks say they never shared secrets. They believe the firings are tied to their push for more AI transparency. OpenAI told The Wall Street Journal the decision had nothing to do with their safety concerns.

The probe began after OpenAI suspected they shared model architecture details with an outside researcher. Investigators found disappearing chats on their phones, and all three were fired a week later. The researchers say the company never told them in writing which data was mishandled.

The big question is whether outsiders should be allowed to check how AI models think. The three once co-wrote a paper calling to keep AI reasoning open for review. An internal probe started after reports said OpenAI's new model Astra shows less of its reasoning.

A new audit found many security holes in OpenAI's systems, but nearly half of the report was redacted. The three researchers have written to the board, asking for closer work with outside safety groups.