Okta Unveils ‘AI Agents’ Platform as 88% of Firms Report Agent Security Incidents

Okta has introduced a new framework and platform aimed at securing enterprise use of AI agents, as organizations grapple with rising risks tied to autonomous systems.

The company said its new offering, Okta for AI Agents, is designed to help businesses manage and control AI agents by answering three key questions: where agents are operating, what systems they can access, and what actions they are allowed to perform.

The launch comes as AI agents evolve rapidly and begin operating with greater autonomy across enterprise systems. Okta cited research indicating that 88% of organizations have experienced suspected or confirmed security incidents involving AI agents, yet only 22% treat them as independent identities within security frameworks.

Ric Smith said traditional security models are struggling to keep pace with the speed and complexity of AI agents, which can execute tasks, access systems and make decisions with minimal human oversight.

The Okta platform aims to address this gap by introducing identity-based controls for AI agents, treating them as “first-class” non-human identities within enterprise environments.

The system includes tools to discover both sanctioned and “shadow” AI agents, including those deployed without IT approval, and register them within a centralized directory.

Okta is also expanding its integration network—currently spanning more than 8,000 applications—to support AI agent platforms such as Boomi, DataRobot and Google’s Vertex AI.

To control what agents can access, the platform introduces a centralized gateway that governs interactions between AI agents and enterprise systems, including applications, APIs and databases. It also includes credential vaulting and access management features designed to enforce least-privilege policies.

A key feature is a “universal logout” mechanism, effectively a kill switch that allows organizations to instantly revoke an agent’s access across systems if suspicious behavior is detected.

The platform also provides logging and monitoring capabilities, enabling organizations to track agent activity and integrate it with existing security tools such as SIEM systems.

The announcement reflects a broader shift in enterprise security as AI agents move beyond experimental use into production environments, where they can execute complex workflows and interact with critical systems.

Okta said the platform will be generally available from April 30, 2026, as companies seek new approaches to secure increasingly autonomous AI-driven operations.