Agent identities are the new service accounts
Part of my job is running access reviews. Human accounts are the easy part: there is an HR system, a manager, a leaver process. The hard part has always been the service accounts, and AI agents have made that part bigger. An agent does not act with its own authority. It acts with whatever identity it was given. In practice that is an Entra app registration or service principal, a managed…
The article discusses the challenge of managing service accounts and AI agents, which have made the task of managing access reviews more complex. Service accounts, such as Entra app registrations or managed identities, are created and managed by teams running pilots rather than by the people who run identity. The three main patterns observed in these service accounts are a lack of ownership, no recertification, and difficulty in quickly stopping these identities when something goes wrong.
The OWASP Top 10 for Agentic Applications for 2026 includes Identity & Privilege Abuse as its own category, highlighting the importance of addressing identity issues. Microsoft's governance guidance for agent identities aims to provide extended controls, including Conditional Access, Identity Protection, governance, and audit logs. However, this solution still lacks coverage for other service accounts like AWS roles or GitHub Apps.
The article identifies several failure modes to look for during reviews, including ownerless identities with tenant-wide permissions, roles with unrestricted actions, outdated credentials, and identities created by individuals who have since left the organization. It also highlights the importance of recertification on a regular schedule, as well as an off-switch to quickly stop a compromised identity.
The author presents "agent-off-switch" (aos) as a tool designed to help with these reviews. aos reads exports from Microsoft 365, AWS, and GitHub, and provides a ranking of findings based on their potential blast radius. Based on a demo tenant, the article outlines various findings and suggests actions to mitigate these risks, such as disabling the offending identity, removing grants and credentials, and verifying the audit log.
By implementing these controls, organizations can better manage their service accounts and AI agents, reducing the risk of unauthorized access and potential damage.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
