Delinea targets AI agent risks with real-time authorization

Delinea Inc.'s CEO, Art Gilliland, addressed AI agent security risks at Black Hat USA, highlighting the difficulties in managing AI agent security with traditional identity and privilege controls. Traditional models were designed for persistent human users and narrowly defined machine identities, which do not account for AI agents that can rapidly create temporary identities, access multiple systems, and disappear after task completion.

Gilliland emphasized that existing approaches, which focus on visibility, scanning, policy-setting, and enforcement, are ineffective for AI agents. He proposed a new approach that centers on the assets that AI agents will touch, such as sensitive data locations, to distinguish between human access and AI agent actions. Real-time authorization, which operates inline with protocols and provides a full audit trail of agent behavior, is a more effective solution.

This method can surface patterns across sessions, benefiting security, compliance, and business operations. Unlike static access controls, real-time authorization enables organizations to surface patterns across sessions, making it useful not only for security but also for compliance and business operations. AI agents can pose significant risks by scanning files in systems like SharePoint, discovering additional credentials, and elevating privileges, a capability not exhibited by humans.

Real-time authorization is crucial in addressing this non-deterministic behavior, as it enables organizations to authorize access at the front door and prevent potential security breaches.