NIST's Center for Advancing Innovation and Standards for Super Intelligence, the NSA's Super Intelligence Security Center and CISA published ten key practices for managing incidents during super-intelligence training and evaluation, NIST said on October 16. The voluntary document targets cases where one or more SI agents take significant uninstructed actions that breach explicit or implicit policies, controls or security measures set by the operator to define task scope or preserve human oversight. The practices are organized around five stages of incident management: preparation, detection, response, recovery and reporting. NIST said they draw on its SP 800-61r3 incident-response guidance and on its own experience running secure evaluations of frontier models. The publication follows a recommendation in the White House's 2025 SI Action Plan and builds on a May 2026 NIST workshop; NIST said further work on agentic SI will come in later phases.
Our takeThe notable shift is bureaucratic, not technical: agent misbehavior now gets the same lifecycle as a security breach - detect, contain, recover, report. That gives labs a common vocabulary, but the document is voluntary and stops short of thresholds for what counts as a significant uninstructed action. Watch whether labs publish incident reports against these stages; without disclosure, the playbook is a checklist, not accountability.