Okta has introduced a multivendor reference architecture for agent runtime security through the newly formed Blueprint Alliance. The initiative aims to assist enterprises in managing the security of AI agents as they transition into production environments by providing a structured framework for monitoring and risk response.
The architecture evaluates agent security based on four key criteria: agent location, authorized capabilities, current activities, and appropriate response protocols. Okta integrates its identity signals with endpoint and network telemetry to detect anomalies in real-time by comparing agent connections against authorized system access.
The platform includes capabilities to deactivate flagged agents to block new sessions. Okta plans to further expand these features at the Agent Gateway to enable the revocation of active tokens and sessions, allowing administrators to redirect agents that exceed their intended operational boundaries.
