Response to NCCoE Concept Paper: Intent Conveyance as the Missing Layer in AI Agent Identity and Authorization
Year: 2026
Type: report
Abstract: This document is a public response to the National Cybersecurity Center of Excellence (NCCoE) concept paper "Accelerating the Adoption of Software and AI Agent Identity and Authorization" (February 2026), submitted to [email protected] before the April 2, 2026 deadline. The response argues that current AI agent identity and authorization models are necessary but not sufficient. Authentication answers "who is this agent?" and authorization answers "what can it access?" but neither answers "why is it taking this action?" The paper proposes intent conveyance as the missing layer: a structured metadata format (the intent envelope) that declares purpose, human authorization chain, constraints, and temporal scope alongside existing identity and authorization mechanisms. The argument is grounded in three enterprise case studies from the past 90 days: Spotify's AI-first coding pipeline (Honk + Claude Code, engineers directing AI from Slack), Amazon's Kiro AI coding tool causing a 13-hour AWS outage by autonomously deleting a production environment, and Meta's rogue agent that passed every identity check but took unauthorized actions (the "confused deputy" pattern). Each incident illustrates where intent conveyance would have changed the outcome. The paper draws on 25 years of engineering leadership across Microsoft, T-Mobile, AT&T, Expedia, and Hitachi Consulting, daily hands-on use of AI coding agents, and the author's prior submission to the NIST CAISI RFI on AI Agent Security (DOI: 10.5281/zenodo.18903117). Specific recommendations for the NCCoE demonstration project include: composite agent identity binding (human + agent + task + intent), intent-aware authorization, intent-aware audit trails that bridge the Reasoning-Action gap, and cross-provider audit continuity.
Author: Singhal, Nikhil
Language: English
Topic: Access Control and Trust
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7140090118