Identity lifecycle management was built for people. They have an employment record, a manager, and a departure date. AI agents have none of that.
As autonomous agents multiply inside enterprise environments, the governance model designed for humans develops blind spots that traditional IGA tools can’t detect. The problem isn’t that the tools are bad. It’s that they were built for a world that doesn’t exist anymore.
No joiner event. When a person joins, HR triggers provisioning. Role-based access kicks in. An approval chain runs. The agent equivalent? A developer commits a config file or makes a platform API call. No IGA workflow fires. No access request gets submitted. The agent lands in production with whatever permissions the developer gave it — usually more than needed.
No mover event. People change roles? HR updates propagate and entitlements recalculate. Agents change scope constantly — new APIs, new data sources, new environments — and none of that generates a mover signal. The IGA platform has no idea the agent’s access surface just expanded.
No leaver event. When someone leaves, HR triggers deprovisioning. When an agent’s workload is retired, nothing fires. Those credentials stay in secrets managers, environment variables, and OAuth servers. Valid. Open.
Runtime scope expansion. Role-based access control works because human job functions are predictable. Agents don’t have fixed boundaries. An agent built to summarize documents might end up querying APIs, writing to storage, and chaining across systems to complete a task. The access surface expands at runtime, driven by the agent’s objective — not by any governance decision made in advance.
What IGA actually sees. When an IGA platform encounters an agent, it sees a service account with an API key. It checks credential age and notes whether the account appeared in the last review. What it misses is that this account is making authorization decisions and traversing application boundaries autonomously.
The fix isn’t simple. It requires treating agents as autonomous principals with their own lifecycle events — provisioning, review, deprovisioning — mapped into the governance layer. Most organizations haven’t started.
