OpenAI fires three safety researchers, who then write to its board

OpenAI has parted ways with three researchers on its safety team, as The Wall Street Journal reported on 1 October and TechCrunch relayed. The company says an internal investigation confirmed they mishandled sensitive information outside its procedures. According to the paper, they allegedly shared it with an outside AI safety organisation.

On 8 October, according to the Anadolu agency, the same paper reported on a letter from the three, Jasmine Wang, Tomek Korbak and Mikita Balesni, to OpenAI's board. They ask it to keep chain-of-thought monitoring, which reviews the models' written reasoning, and to work with independent auditors. They warn that nobody yet knows how to deploy models safely when they can't be monitored. They deny having dealt with outside parties beyond what their jobs required.