SailPoint’s reported unveiling of AI identity agents and runtime controls at its Navigate event highlights a shift in identity security: organisations are beginning to treat software agents as actors that need governed access, rather than as ordinary application features. For identity governance and administration (IGA), the central question is how to make an agent’s authority visible, limited and accountable throughout its operational life.
Agents complicate familiar controls
Traditional identity processes are built around people and established service accounts. AI agents can behave differently: they may execute tasks across multiple systems, call tools dynamically and pass work to other agents. If an agent inherits a user’s broad permissions or relies on a long-lived credential, its effective access can exceed what the business intended. Investigating a consequential action may also be difficult if logs identify only the initiating user or a generic technical account.
Runtime controls address part of this problem by evaluating activity as it happens. They may help enforce policy at the point of access, but they do not replace lifecycle governance. Organisations still need to know who created an agent, who sponsors it, what purpose it serves, which systems it can reach and when its authority should expire.
Applying IGA disciplines
Agent identities should have accountable owners and a recorded business purpose. Access should be granted through approved workflows, scoped to the task and separated by environment where appropriate. A development agent should not automatically gain production privileges. Where agents use delegated access, the organisation should preserve the distinction between the human or service principal that initiated work and the agent that performed each action.
Least privilege must be practical, not static. Teams can begin with a narrowly scoped permission set, observe required actions and adjust through controlled approvals. High-impact capabilities, such as changing entitlements, exporting sensitive data or creating additional credentials, merit stronger restrictions and explicit policy checks.
IGA reviews can include agent owners, permissions and activity evidence. Reviewers need context about what the agent does and which actions are expected. Certification should cover the agent’s effective permissions, including indirect access through tools and delegated credentials, rather than just the nominal role assigned in a registry.
Evidence and response
Security teams should be able to trace an action to an agent identity, its owner, the authorising request and the resources accessed. Logs should support investigation without exposing secrets. Revocation must be tested: disabling an agent should invalidate credentials, tokens and delegated sessions, not merely hide its registration record.
Runtime enforcement can help detect anomalous actions, constrain calls and interrupt misuse. Governance programmes should connect those events to incident response and periodic access review. They should also define how to handle model updates, changes in tool access and agents that are copied or reconfigured.
Organisations assessing new agent controls should ask how identities are registered, how permissions are granted and reviewed, and whether every action is attributable. Agentic systems increase the number and speed of access decisions. Strong IGA provides the ownership, lifecycle and evidence needed to keep those decisions governable.