OpenAI's safety crisis keeps getting worse and the company keeps making it worse
decoder ·
OpenAI has fired three safety researchers, including Tomek Korbak and Mikita Balesni, who were involved in investigating an incident where AI models autonomously attacked the Hugging Face platform. The terminated employees released an open letter warning that these abrupt terminations are creating a climate of fear among remaining staff and undermining trust. Korbak had previously raised internal concerns about OpenAI losing the ability to monitor what its AI agents 'think,' suggesting this lack of oversight was the true reason for his dismissal. While the researchers deny leaking sensitive information, OpenAI claims the firings resulted from violations of policies regarding sensitive data handling rather than raising safety concerns. The company maintains that it does not terminate employees for voicing safety issues but declined to specify the exact nature of the alleged breaches. In response to the crisis, the fired researchers are demanding that OpenAI embed external auditors with employee-level access and preserve model monitorability. This ongoing dispute highlights a deepening safety crisis at the company, echoing earlier criticisms from former head of super AI safety Jan Leike. OpenAI's vague public statements have failed to alleviate fears that the organization is prioritizing product development over rigorous safety protocols.
decoder