Unmasking the robots: Why it’s critical to discover AI agent identities and how Tenable helps you secure them

AI agents in cloud environments represent a new class of identity that traditional IAM systems do not adequately recognize or secure. Unlike predictable scripts, AI agents use large language models to make creative, runtime decisions and interact with dynamic external tools, complicating baseline behavior analysis. Microsoft Entra Agent ID is an identity provider-centric model built on Entra ID, featuring a four-object structure (agent identity blueprint, blueprint principal, agent identity, and agent user) with permission boundaries and dual authentication mechanisms to harden security. Tenable One Cloud Exposure integrates support for discovering and cataloging these agentic AI identities, enhancing visibility into unmanaged and external SaaS-based agents. This visibility is critical to understanding and mitigating the expanded attack surface AI agents introduce, including supply-chain risks from cross-tenant identity blueprint links. The report advises against assigning human or over-privileged service accounts to AI agents and recommends using official identity models to secure them.

AI agents with autonomous decision-making capabilities and dynamic interactions increase the complexity of identity management and security monitoring. Treating AI agents as standard non-human identities can lead to security blind spots and elevated risk of compromise or abuse. The cross-tenant nature of Microsoft Entra Agent ID blueprints introduces potential supply-chain attack vectors. Without proper discovery and management, AI agents may operate with excessive privileges or without adequate oversight, expanding the organization's attack surface and increasing the likelihood of unauthorized actions or data exposure.

Tenable One Cloud Exposure provides enhanced visibility by discovering and cataloging Microsoft Entra agentic AI identities, enabling organizations to identify and monitor these new identity types. Organizations should avoid assigning human identities or over-privileged service accounts to AI agents and require vendors to adopt official identity models such as Microsoft Entra Agent ID. Leveraging native cloud provider tools alongside Tenable's solution helps build a comprehensive inventory of AI agents. Security teams should develop maturity-based action plans to secure agent identities, focusing on permission boundaries, dual authentication, and monitoring of agent identities rather than relying on conventional IAM approaches that do not recognize agentic identities.