Accelerating the Adoption of Software and AI Agent Identity and Authorization
A 12-page technical response proposing Attested Governance Artifacts as the cryptographic proof for AI agent identity binding, continuous runtime authorization, and offline-verifiable audit evidence.
Download PDF· 12 pagesWhat this submission covers
This submission responds to the NCCoE concept paper on AI Agent Identity and Authorization by mapping Attested Governance Artifacts across six technical categories. It demonstrates how sealed policy artifacts, signed governance receipts, and tamper-evident continuity chains address identity binding, runtime authorization, and non-repudiation for autonomous AI agent systems.
Six technical categories
Use Cases & Scenarios
Threat landscape for autonomous AI agents including credential theft, phantom execution, and retroactive fabrication attacks.
Identification & Binding
Cryptographic identity binding through sealed policy artifacts with Ed25519 signatures and SHA-256 continuity chains.
Authentication Mechanisms
Runtime authentication via integrity measurements at every tool call. Session-initiation checks alone leave a wide exposure window.
Authorization Frameworks
Sealed policy artifacts define the permitted operations; the gateway renders and seals a permit-or-deny decision, and effecting it is a per-deployment integration point.
Auditing & Non-Repudiation
Tamper-evident continuity chains and offline-verifiable evidence bundles built for air-gapped audit environments.
Prompt Injection Defenses
Behavioral drift detection through runtime measurement of tool-call patterns against sealed baseline profiles.
Proposed Lab Demonstration
A four-phase demonstration protocol for the NCCoE lab environment covering artifact creation and sealing, runtime measurement and sealed decisions with drift detection, evidence bundle generation, and offline verification by anyone holding the key.