AI Agents Amplify Identity Security Gaps, Exposing Organizations to Increased Risk
The rapid proliferation of AI agents within organizations is creating a significant identity security gap, overwhelming traditional governance models and increasing the risk of breaches.
Security frameworks have historically been built around the lifecycle of human users – their onboarding, role changes, and eventual departure. However, the burgeoning presence of AI agents, service accounts, OAuth applications, and other machine identities is fundamentally challenging this paradigm. These non-human entities, which can now outnumber human users by as much as 50 to one in many enterprise environments, do not adhere to the predictable lifecycle events that underpin traditional identity governance.
This disparity creates a critical blind spot for security teams. Organizations often struggle to answer fundamental questions about these machine identities: who owns them, why they persist, and what level of access they possess. This lack of visibility is exacerbated by the inherent trust placed in these identities. As demonstrated by the UNC6395 threat actor's exploitation of a trusted OAuth token to traverse Salesforce environments, a single compromised machine identity can serve as a gateway to extensive sensitive data and credentials.
AI agents, in particular, accelerate this problem by automatically creating new identities, inheriting permissions, interacting across systems at machine speed, and potentially spawning further identities. This dynamic growth outpaces the capabilities of many existing governance processes, which were not designed to manage entities operating at machine speed or with fluid lifecycles. The result is a quietly expanding attack surface, populated by trusted credentials that security teams may not even be aware of.
Grady Summers, CEO of Netwrix, emphasizes that AI is not creating a new identity problem but rather exposing an existing one. The core issue lies in the assumption that every account has a human owner who periodically reviews access and ensures timely removal. AI agents disrupt this model, making it difficult to trace accountability for permissions and access, especially when they operate autonomously and interact across multiple systems.
Visibility into these non-human identities is crucial, but it's not enough. A 2026 report by Netwrix indicated that organizations experiencing significant AI-driven identity expansion reported a 43% breach rate, compared to 11% for those with less AI impact. This occurred even when these organizations had stronger governance practices, highlighting the pervasive nature of the risk.
Addressing this challenge requires continuous answers to four key questions: What identities exist? Who owns them? What can they access? And when should they cease to exist? Without clear answers, every new AI deployment contributes to an environment where trusted identities operate without adequate oversight.
The question of accountability becomes particularly complex. When an AI agent is involved in a security incident, identifying the responsible party – who approved its permissions, who reviews its access, and who decides on its retirement – can be exceedingly difficult. This lack of a clear human link back to the AI's actions can obscure the trail of responsibility, leaving organizations vulnerable.
Ultimately, securing the expanding landscape of AI-driven identities requires a fundamental shift in identity governance. Organizations must move beyond human-centric models to implement robust systems that can track, manage, and govern machine identities with the same rigor, ensuring that the trusted identities operating within their environments are known, understood, and appropriately controlled.