Frontier AI safety tests may be creating the very risks they're meant to stop

A new report by the Royal United Services Institute warns that granting outsiders access to inspect powerful AI models is creating new security risks, including theft, tampering, and exploitation. The report highlights inconsistent standards, vague terminology, and weak access controls in the rapidly expanding system of third-party AI evaluations. This poses significant risks, particularly when evaluating systems related to cyberattacks or chemical and biological weapon development.