An Iranian-linked group used artificial-intelligence models to identify vulnerabilities in the ship-born communications of U.S. Navy warships, according to an Anthropic national-security report