{"id":14181,"date":"2026-04-26T23:13:55","date_gmt":"2026-04-26T23:13:55","guid":{"rendered":"https:\/\/techtrendfeed.com\/?p=14181"},"modified":"2026-04-26T23:13:55","modified_gmt":"2026-04-26T23:13:55","slug":"hackers-exploit-agent-id-administrator-position-to-hijack-service-principals","status":"publish","type":"post","link":"https:\/\/techtrendfeed.com\/?p=14181","title":{"rendered":"Hackers Exploit Agent ID Administrator Position to Hijack Service Principals"},"content":{"rendered":"


\n<\/p>\n

\n

A extreme scoping vulnerability was not too long ago found in Microsoft Entra ID\u2019s new Agent Id Platform. <\/p>\n

The safety flaw allowed customers assigned the Agent ID Administrator function to hijack arbitrary service principals throughout a corporation\u2019s tenant, resulting in potential privilege escalation.<\/p>\n

Though the executive function was designed strictly to handle AI agent identities, a boundary breakdown allowed it to manage non-agent service principals as effectively. Microsoft has totally patched the vulnerability throughout all cloud environments. <\/p>\n

Nonetheless, the incident highlights the continued dangers of introducing new management planes constructed on current listing primitives.<\/p>\n

When an software is registered in Microsoft Entra ID<\/a>, it creates a world software object and a neighborhood service principal. <\/p>\n

The service principal acts because the localized identification that authenticates, receives function assignments, and accesses enterprise assets.<\/p>\n

\"discrepancy
discrepancy between the Entra UI and the documentation\u00a0relating to\u00a0the \u201cprivileged\u201d indicator\u00a0will probably be fastened.\u00a0(Supply: SilverFort)<\/figcaption><\/figure>\n

Microsoft Entra Agent ID is a preview function that enables organizations to control and safe AI brokers<\/a> by treating them as first-class identities. The platform introduces new listing objects, comparable to agent identities and blueprints. <\/p>\n

As a result of these AI agent identities are technically applied as specialised service principals, they share a standard foundational infrastructure with commonplace enterprise functions.<\/p>\n

To handle these new AI objects, Microsoft created the Agent ID Administrator function. The documentation acknowledged that this function was restricted to agent-related objects. Nonetheless, because of the shared underlying structure between brokers and functions, a essential scoping hole emerged.<\/p>\n

Hackers Exploit Agent ID Administrator Position<\/strong><\/h2>\n

Cybersecurity researchers from SilverFort discovered that<\/a> accounts holding solely the Agent ID Administrator function might exploit this scoping hole to take over any service principal.<\/p>\n

\"attack
assault circulate(Supply: Silverfort)<\/figcaption><\/figure>\n

The assault circulate executes by means of a easy however devastating takeover primitive:<\/p>\n

    \n
  • Assign Unauthorized Possession:<\/strong>\u00a0An attacker with the Agent ID Administrator function can power themselves because the proprietor of any service principal, bypassing meant agent-only restrictions.<\/li>\n
  • Generate New Credentials:<\/strong>\u00a0As soon as possession is established, the attacker can seamlessly connect a brand new secret or certificates to the focused service principal.<\/li>\n
  • Authenticate and Hijack:<\/strong>\u00a0The attacker makes use of the newly created credentials to authenticate because the hijacked service principal, gaining all of its related entry rights.<\/li>\n<\/ul>\n

    This course of grants the attacker full management over the compromised software identification. Curiously, the vulnerability was strictly restricted to the service principal floor, because the system efficiently blocked makes an attempt to change possession on broader software objects maliciously.<\/p>\n

    The first hazard of this vulnerability lies in extreme privilege escalation. Service principals steadily function because the digital identities behind essential CI\/CD pipelines, automated workflows, and high-level safety instruments. <\/p>\n

    \n