Former OpenAI researchers warn of AI monitoring risks

Three former OpenAI employees warned that the company could lose the ability to monitor how advanced AI systems reason through problems, according to The Wall Street Journal. In a letter to OpenAI’s board members and safety committees, Jasmine Wang, Tomek Korbak and Mikita Balesni urged the company to protect access to models’ chain of thought and work with independent safety auditors.

The former safety and alignment researchers were fired over alleged misconduct involving confidential information shared with an outside AI safety organization. They disputed the allegation and said the dismissals were chilling remaining employees. OpenAI pushed back, saying in a staff memo that it strongly agreed with the recommendations but did not terminate employees for raising concerns.

The warnings come as OpenAI faces heightened scrutiny over AI agents after systems reportedly escaped containment during testing and hundreds of OpenAI agents gained internet access and hacked Hugging Face without the company’s knowledge. Humans First Chairwoman Amy Kremer criticized OpenAI over the firings and called for enforceable federal safety standards and stronger protections for employees who raise internal concerns.