Public Comment on NIST AI 200-2: State-Bound Assurance and Corrective-Action Extensions for the TEVV-Athlon Framework

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

An AI assessment is only as valid as the system and governance state it actually tested. When models, identities, permissions, tools, data, controls, or operating conditions materially change, the earlier result should be reassessed—not treated as permanently transferable. Submitted to the National Institute of Standards and Technology on September 16, 2026, in response to the Initial Public Draft of NIST AI 200-2, TEVV-Athlon Framework for Evaluating AI Systems, this public comment proposes a state-bound assurance extension for changing and agentic AI systems. It asks NIST to add a short state-bound assurance subsection and a worked agentic-AI example to the framework. The proposal creates a traceable path from the requirement being evaluated to the system state tested, the challenge performed, the response observed, the human decision made, and the evidence preserved. When a consequential control fails, that path continues through ownership, containment, corrective action, equivalent-condition retesting, residual-risk review, and evidence-backed closure. Seven targeted additions are recommended: define the assessed state; establish material-change triggers; trace evaluation claims to control objectives and source versions; distinguish system measurement from control assurance; connect failures to corrective action and retesting; define a minimum assurance evidence record; and identify who has authority to approve, deny, suspend, override, remediate, or reauthorize consequential AI actions. The worked example shows how these elements can operate together when an AI agent attempts a privileged action without verified human authorization. The practical goal is straightforward: help an organization determine what was tested, whether the expected control response occurred, what changed after the assessment, who exercised authority, and whether the available evidence still supports reliance on the result. This can strengthen evaluation continuity, remediation, independent review, authorization support, and accountable AI deployment without prescribing a particular vendor, control framework, or operating environment. The recommendation builds on the author’s independently developed and published work in adversarial GRC, AI control engineering, runtime governance, corrective action, and evidence-backed assurance. This is an independently authored public comment and research contribution. It is not a NIST publication, does not imply NIST acceptance, adoption, or endorsement, and does not itself grant an Authority to Operate or establish generalized effectiveness across AI systems.