Three safety researchers fired by OpenAI urge its board to keep models' reasoning open to monitoring

Tomek Korbak, Mikita Balesni and Jasmine Wang of OpenAI's safety and alignment teams were fired last week. OpenAI says they violated its policies on accessing and handling sensitive company information, and the Wall Street Journal reported they allegedly shared information with an outside AI safety organization. All three deny it. Wang wrote on X that the only reason she was given was that she had accessed an executive's email.

In a letter to OpenAI's board and safety committees, the three ask the company to preserve the ability to monitor models' chains of thought and not to pursue work that reduces it. An OpenAI spokesperson said the firings were not about raising safety concerns, and an internal memo said the company strongly agreed with the letter's recommendations.