Agentic AI’s Identity Problem Leaves Organizations Vulnerable to Attackers
A growing concern in the cybersecurity community is the lack of proper identity management for artificial intelligence (AI) agents, which are increasingly being used across various industries. These digital actors authenticate, receive permissions, and take actions on behalf of organizations, often with credentials and access that are not fully understood or controlled by security teams.
The issue lies in the fact that AI agents behave like humans, interpreting goals and choosing paths to achieve them, but scale like software, processing at machine speed. This autonomy, combined with their ability to be created quickly and embedded into various systems, makes it challenging for traditional identity programs to keep up. As a result, these agents often possess overly broad access, which can lead to significant security risks if not properly managed.
Traditional least privilege models, which are commonly used in human identity management, fail to scale when applied to agentic AI. Least privilege typically involves granting the minimum static permissions required for a role or function, but an agent’s needs can vary depending on its goal, data involvement, and environment. For example, a support agent summarizing a ticket requires different access than one that can issue refunds or modify customer records.
The lack of proper identity management for AI agents creates a perfect storm for attackers to exploit. Many organizations are unaware of the existence of these agents, which operate in the shadows, using credentials and API tokens without being inventoried or reviewed. This makes it difficult for security teams to understand the scope of the blast radius and hold anyone accountable when an agent makes a harmful decision or is abused by an attacker.
To address this issue, organizations need to adopt intent-based policies that are contextual, time-bound, and continuously evaluated. However, this requires significant changes in how enterprises operate today. AI agents must be governed at scale without sacrificing speed, which can be achieved through automated tools that discover every agent, map risky access, and enforce intent-based policies.
In conclusion, the identity problem facing agentic AI is a pressing concern for organizations that rely on these digital actors. By understanding the nature of this issue and adopting new approaches to identity management, security teams can mitigate the risks associated with AI agents and ensure the secure adoption of these technologies.
Source: Bleeping Computer — 2026-06-29