AI Agents Bypass Identity Controls, Posing Security Risks

In environments examined for the 2026 State of Agent Security Report, approximately 1,280 third-party products now integrate artificial intelligence capabilities. Of these, around 282 are protected by single sign-on (SSO) authentication. However, the remaining 1,000 products operate outside the purview of traditional identity infrastructure by default. This situation arises not from intentional concealment, but from the inherent limitations of identity stacks, which can only govern and monitor entities that authenticate through them. Most AI agents do not authenticate via these standard systems, leaving a significant security gap.

The report highlights that this gap is the most apparent security challenge associated with the proliferation of AI agents. These agents, designed to perform tasks autonomously or semi-autonomously, often interact with various systems and data sources without explicit user authentication for each interaction. This lack of visibility means that security teams cannot effectively monitor, manage, or revoke access for these agents, even if they are compromised or acting maliciously. The study specifically points out that the security frameworks currently in place are largely designed for AI that users consciously select and integrate, failing to account for the pervasive and often invisible nature of third-party AI agents.

This invisibility extends to the potential for these agents to introduce new attack vectors or to be exploited by malicious actors. Without proper authentication and authorization mechanisms, a compromised AI agent could potentially access sensitive data, execute unauthorized commands, or disrupt operations without detection. The report implies that the current security paradigm, which relies heavily on user-centric identity management, is insufficient to address the unique challenges posed by autonomous AI agents operating within organizational networks. The sheer volume of these agents, with over a thousand operating without standard identity oversight, underscores the scale of the problem.

The implications of this security gap are far-reaching. Organizations are increasingly embedding AI into their workflows and products, often through third-party solutions. If these solutions are not adequately secured and monitored, they represent a significant blind spot for cybersecurity professionals. The report suggests a fundamental re-evaluation of how AI agents are managed and secured, moving beyond traditional identity and access management (IAM) principles to develop new strategies that can provide visibility and control over these autonomous entities. The challenge lies in securing AI that is not directly chosen or managed by the end-user, but rather embedded within the tools and services they utilize.