Information security
fromtheregister
14 hours agoFrontier AI safety tests may be creating the very risks they're meant to stop
Third-party AI evaluations require outsider access, but inconsistent standards and weak controls create new risks of theft, tampering, espionage, and abuse.