AI risk reclassification should respond to material changes in how an AI system is used, not only to new model releases. See how changes in purpose, population, data, decision consequence and automation can trigger a fresh assessment of whether the existing risk classification and controls still fit.
Microsoft and NVIDIA publish joint guidance grouping open-weight model risks into five categories and mapping selected risks to evaluations, with agent and runtime guidance planned.
NIST, the NSA and CISA issue ten voluntary incident-response practices for superintelligence training and evaluation, addressing unexpected agent behaviour, evidence preservation, containment and reporting.
AI decision audit evidence connects the records behind one consequential AI outcome. See how decision identity, source state, model and policy versions, human intervention and committed business action can form a reconstructable evidence packet without duplicating every underlying log.
AI policy exception management keeps temporary governance waivers from becoming permanent control gaps. See how scope, ownership, compensating controls, expiry and renewal can remain machine-visible while the underlying policy stays intact.
Multi system AI decision accountability prevents business outcomes from falling between technical owners. See how enterprises can assign a named owner to each decision class while data, model, policy, agent and transaction teams remain accountable for their individual controls.
AI agent decision rights governance defines which business judgments an agent may make before technical permissions implement that authority. See how discretion, exceptions, consequence, reversibility and escalation can determine where autonomous decision-making ends and human authority remains.
Board AI governance control view turns an AI inventory into decision-ready oversight. See how business consequence, control condition, evidence, exceptions, residual risk, trends and management accountability help boards focus on the AI exposures that actually require governance judgment.
Continuous AI assurance framework keeps approval evidence current as models, tools, data, suppliers and operating conditions change. See how scheduled reassessment, material-change triggers and production monitoring can refresh assurance throughout the lifecycle while annual reviews become portfolio-level summaries.
Third party AI risk models APIs tools extends vendor governance beyond the supplier register. See how foundation models, external APIs, embedded AI, connectors and subprocessors can change data flows, authority and operational dependencies even when the primary vendor relationship stays the same.