Okta outlines controls for AI agent insider risk at Oktane

Okta is addressing the insider risk created by AI agents through discovery, authorisation and runtime controls, according to Krista Case, principal analyst at theCUBE Research. Speaking at Okta’s Oktane event, Case said agents that act across applications and delegate work to other agents can change systems faster than security teams can review them.

Case highlighted a gap in the common kill switch response: revoking an agent’s authority stops future actions but does not undo changes already made or downstream actions it triggered. She said agents need clear human ownership and permissions tailored to their tasks rather than inheriting all of a user’s access.

The scale of the problem is already visible. Case cited a financial asset management company that discovered about 13,000 agents in its environment but considered only 1,000 valid, suggesting many are created outside established IT processes. Okta’s Blueprint Alliance aims to establish a shared architecture for access, delegation and monitoring across vendors.