800-plus 'potential vulnerabilities and biases' may afflict AI-infused medical services

More than 800 “potential vulnerabilities and biases” were uncovered by a Pentagon effort to spot problems with using large language models in military medical services, officials said Thursday.The Chief Digital and Artificial Intelligence Office, or CDAO, said the initiative was conducted through its Crowdsourced AI Red-Teaming Assurance Program, with help from the Program Executive Office, Defense Healthcare Management Systems, and the Defense Health Agency.