Background: The rise of agentic artificial intelligence (AI) — autonomous, goal-driven software entities that act on behalf of users or organizations — introduces novel identity, access, and trust challenges for modern networks. Traditional perimeter-based models of security are ill-suited for environments where autonomous agents, dynamic workloads, and decentralized identities interact across hybrid cloud, on-premises, and edge infrastructures (Gilman & Barth, 2017; Department of Defense CIO, 2007). Recent proposals emphasize integrating Zero Trust principles with intent-aware identity management to protect AI workloads and agentic behaviors (Hasan, 2024; Achanta, 2025; Kumar, 2023).
Objective: This research article proposes a comprehensive, publication-ready architecture — an Intent-Aware Zero-Trust Identity Architecture (IAZTIA) — that unifies human and machine access, supports agentic AI, and enforces continuous, intent-based policy decisions while accounting for non-stationarity, noisy labels, and adversarial behaviors in telemetry and identity signals (Anderson & McGrew, 2017).
Methods: The architecture synthesizes established standards and operational practices including FIPS 199 security categorization, Cloud Security Alliance Secure Device Posture and SDP concepts, hardware asset management, SPIFFE/SPIRE identity federation mechanisms, decentralized identifiers (DIDs), and intent-based network virtualization principles (NIST FIPS 199, 2004; CSA-SDP, 2015; HWAM, 2015; CNCF/SPIFFE, 2024; W3C, 2023; IBNVN, 2013). We describe a layered methodology: identity provenance and binding, intent extraction and semantic normalization, continuous policy evaluation under Zero Trust, telemetry validation and robust learning for noisy labels, and governance controls for accountability and audit. Design choices are grounded in threat and risk taxonomies developed for agentic AI (OWASP, 2024; OWASP Agent Risk, 2024; Syros et al., 2025).
Results: The IAZTIA design presents: (1) identity constructs that bind human, device, and agentic AI identities using short-lived cryptographic credentials and verifiable DIDs; (2) an intent model capturing goals, constraints, and permitted action templates for agents; (3) a policy decision and enforcement fabric leveraging SPIFFE/SPIRE and SDP-aligned micro-segmentation; (4) robust telemetry pipelines applying practices from malware traffic and noisy label research to maintain policy fidelity (Anderson & McGrew, 2017); and (5) governance controls for role separation, lifecycle management, and incident forensics (Hassan, 2025; Bhushan et al., 2025). We further provide attack scenarios and mitigations, and propose measurable metrics for resilience and trustworthiness.
Conclusions: IAZTIA advances the state of practice by explicitly combining intent semantics with Zero Trust identity controls for agentic AI, enabling continuous, contextual access decisions while providing auditability and governance. The architecture addresses known challenges — identity sprawl, telemetry poisoning, credential misuse, and non-stationary behavior of agents — and outlines a path for operational adoption integrating standards and cloud-native identity primitives (Cohen et al., 2013; Gilman & Barth, 2017; W3C, 2023).