Identity Lifecycle Management Wasn't Built for AI Agents
13:30 · July 2, 2026 · Hacker News AI Section
Identity lifecycle management was architected around a person with an employment record, a manager, and a departure date. AI agents have none of those. As autonomous principals proliferate across enterprise environments, the governance model built for humans develops structural blind spots that traditional IGA tools weren't designed to detect. This guide covers where that model breaks, what it
Summary
Identity lifecycle management systems were built around a human principal whose organizational status is tracked through HR-driven events. Provisioning, role changes, and deprovisioning are triggered by records in systems such as Workday or SAP SuccessFactors, which feed joiner-mover-leaver signals into IGA platforms. These signals allow deterministic mapping of entitlements, periodic access certification by managers, and reliable offboarding when employment ends.
AI agents bypass this model entirely. They are instantiated through deployment pipelines, Terraform runs, or orchestration frameworks rather than HR records, arriving with credentials that are often created inline and never registered as governed identities. Because agents lack employment attributes, managers, or fixed departure dates, none of the canonical lifecycle events are generated. The IGA platform therefore sees only a static service account or OAuth client, while the actual principal operates with runtime autonomy.
This autonomy produces expanding access surfaces that traditional controls cannot observe. An agent may chain tool calls or retrieve data through RAG patterns that reach APIs and storage systems outside its original scope. In multi-agent setups, orchestrators can spawn sub-agents and pass credentials across execution contexts, creating parallel instances whose permissions are never reconciled against a single identity record. Access reviews receive no updated attributes and therefore generate no attestation tasks.
Offboarding is equally invisible. When an agent workload is retired, its credentials remain in secrets stores and authorization servers because no termination event reaches the IGA layer. The result is persistent, ungoverned access paths that accumulate across the agent population without triggering any of the compliance or risk controls designed for human identities.
Why it matters
This is highly relevant for security and privacy professionals in the Netherlands as the adoption of autonomous AI agents grows. Proper identity and access management for AI is essential to maintain compliance with EU regulations like the AI Act and GDPR, preventing unauthorized data access and lateral movement.
