A reported 40-fold gap between enterprise AI agent adoption and identity security readiness is a striking way to frame a growing control problem. Whether the precise ratio varies by organisation, the underlying concern is familiar to identity governance teams: new agents can be deployed faster than inventories, approvals and monitoring processes can keep pace.
Adoption is creating identity sprawl
AI agents are not simply another software licence. They may act on behalf of people, connect to multiple services and hold credentials that permit consequential changes. Teams can create agents for pilots without involving IAM, then promote them into production while their access remains poorly documented. The result is a population of non-human identities that may lack owners, defined lifetimes or meaningful oversight.
Conventional security controls can miss this shift. Endpoint tools may see the host, and application logs may record an API call, but neither necessarily explains which agent acted, whose authority it used or whether the access was appropriate for the task. Periodic review alone is insufficient if the number and function of identities change rapidly.
Close the visibility gap first
Identity governance begins with knowing what exists. Organisations need to discover agents and associated service accounts across cloud platforms, SaaS products and development environments. Each record should identify an accountable owner, business purpose, environment, connected tools and permissions. Reconciliation against infrastructure and application data helps surface identities that bypassed formal onboarding.
Classification can make a large inventory actionable. An agent that can read public documentation is not equivalent to one that can alter financial records or administer cloud infrastructure. Risk tiers can guide approval requirements, review frequency and monitoring. This also helps security teams focus scarce attention on identities with broad reach or sensitive capabilities.
Make access lifecycle-based
Identity lifecycle management should cover creation, modification, suspension and retirement of agents. Access should be based on a clear use case and scoped to the minimum resources required. Temporary credentials and just-in-time elevation can reduce standing privilege, while automated expiry prevents pilot permissions from silently becoming permanent.
Governance must account for changing context. When an agent’s owner leaves, its model changes, or a workflow gains a new integration, existing permissions may no longer be appropriate. Those events should trigger reassessment. High-risk actions can require human approval, and policies should prevent agents from granting themselves broader access or creating unmanaged downstream identities.
Measure readiness with evidence
Leaders can track the share of agents with verified owners, the proportion of permissions linked to approved purposes, the time required to revoke access and the number of dormant or excessive identities found. These indicators reveal whether controls are catching up, rather than relying on a headline adoption statistic.
IGA is not a brake on AI. It provides the accountability and access discipline that allow organisations to scale automation without treating every agent as inherently trusted. Closing the gap requires collaboration among IAM, security engineering, data governance and business teams, with identity governance administration embedded in the agent lifecycle rather than added after deployment.