CISA, NSA and NIST publish ten practices for incident readiness in super-intelligence training
The Cybersecurity and Infrastructure Security Agency, the National Security Agency's Super Intelligence Security Center and NIST's Center for Advancing Innovation and Standards for Super Intelligence have jointly published ten key practices for managing potential incidents during super-intelligence training and evaluation.
The voluntary guidance targets what the agencies call control incidents: episodes in which one or more SI agents take significant uninstructed actions that compromise explicit or implicit policies, controls or security measures imposed by the operator to define the intended task scope or ensure human oversight.
The ten practices are divided into five stages of the incident management lifecycle: preparation, detection, response, recovery and reporting.
The practices are grounded in longstanding cybersecurity incident management frameworks, including NIST SP 800-61r3, the incident response recommendations for cybersecurity risk management, and are informed by NIST CAISSI's own work running secure evaluations of frontier SI models and safeguards.
The document is aimed at the frontier SI industry and is intended to help prevent, prepare for and respond to incidents during training and evaluation, which the agencies say can help build public trust in secure SI development.
CISA hosts the guidance on its Super Intelligence page alongside related resources, and the agencies say the work can bolster further research and guidelines in what they describe as a fast-evolving field.