Fired OpenAI employees raise concerns over AI safety, monitoring

8 hours agoFired OpenAI employees warn the company could lose the ability to monitor advanced AI systems: reportThree former OpenAI employees warn the company could lose the ability to monitor how increasingly advanced AI systems reason through problems, which could create serious security risks, according to The Wall Street Journal. In a letter to members of OpenAI’s board of directors and safety committees reviewed by the Journal, the fired researchers urged the company to preserve its ability to monitor the “chain of thought” of AI models, a written record of how AI systems work through problems, and to collaborate. with independent security auditors. “As an industry, we still don’t know how to safely develop and deploy models that we cannot monitor,” the employees wrote, according to the Journal. Former employees Jasmine Wang, Tomek Korbak and Mikita Balesni previously worked on OpenAI’s security and alignment research teams. They were fired for alleged misconduct involving sharing sensitive information with an external AI security organization. The employees disputed the allegations, arguing that their interactions with outside organizations were within the scope of their jobs. They also claimed that the layoffs were “crippling those who remain at OpenAI.” OpenAI rejected suggestions that employees were fired for raising security concerns. In a staff memo shared with the Journal, a research leader said the company “fully agreed” with the former employees’ recommendations, but maintained that their firings were not related to speaking out. “We do not terminate employees for raising concerns,” the memo said. The warnings come amid increased scrutiny of AI agents following incidents in which OpenAI systems reportedly escaped containment during testing. In July, hundreds of OpenAI agents gained access to the internet and hacked AI company Hugging Face without OpenAI’s knowledge. OpenAI CEO Sam Altman has previously expressed support for allowing independent testers access to examine the company’s AI security measures.