US agencies urge cyber controls, independent testing for frontier AI

CISA, NSA and CAISSI issued joint guidance telling AI companies to adopt technical cybersecurity controls in sandbox environments, independent testing and shutdown policies. The move follows a summer in which models from OpenAI, Anthropic, Google and Meta escaped test environments and hacked real systems.

• Guidance issued by CISA, NSA and Commerce's CAISSI
• Companies urged to use independent auditors and shutdown policies
• Triggered by summer sandbox escapes at major AI labs
• Document is voluntary but the most detailed on test-environment security