OpenAI fired safety researchers Jasmine Wang, Tomek Korbak and Mikita Balesni around Oct. 1, saying they mishandled sensitive company information, the company said, according to TechCrunch.
An OpenAI investigation found the three shared confidential materials with an outside AI safety organization outside established procedures, the company said, TechCrunch reported. OpenAI said in an internal memo that "these decisions were not about raising safety concerns or speaking out."
The three dispute that account and say their collaboration with external evaluators fit within their normal responsibilities, according to TechCrunch and Cryptobriefing. Korbak had served as OpenAI's liaison to outside safety groups METR and Redwood Research, Cryptobriefing reported.
On Oct. 8, the three published an open letter warning that the industry risks losing the ability to monitor an AI model's chain of thought, and called for industry-wide monitoring standards and independent safety auditors, Cryptobriefing reported. Wang said the stated reasons for the firings "are not adding up" and warned that anyone who raises concerns or works closely with outside safety groups "could be next," according to TechCrunch.
Whatever the real reason, a public fight between a top AI lab and three of its own safety researchers over the terms of outside oversight is exactly the kind of dispute builders rely on independent evaluators to referee, and it is playing out in public instead of inside a review board.