TL;DR Cayosoft Guardian 7.4 lets you monitor and investigate AI agent identity changes in Entra ID with a plain-language question and roll back what you didn’t approve.
An AI agent that reads email, updates records, or manages users needs an identity to do it, and in most tenants that identity is already there. Some agents use Microsoft Entra Agent ID, Microsoft’s dedicated model for agent identities. Others run as an ordinary app registration with a client secret and API permissions, as a service account, or under the credentials of the person who built them.
That identity requires the same oversight as any other identity in the tenant. The identity determines what the agent can access and do, so when it changes, the agent’s access and capabilities can change with it. Those changes need to be visible.
Now, Cayosoft Guardian makes changes to those identities easier to investigate and reverse. Identity teams can ask what changed in plain language, find the relevant changes across their history, and roll back changes they didn’t approve.
Every AI Agent Has an Identity that Needs Monitoring
Agent identities rarely go through the process that governs human accounts. There’s no hire date, no manager approval, and no offboarding. A developer or a platform creates them, permissions get added as the agent takes on more work, and nobody is assigned to review them.
And if an agent is built to act autonomously, excessive access can be used without a person reviewing or approving each action.
These changes can quietly expand what an agent can do or who controls it:
| Watch for | Why it matters |
|---|---|
| New agent identities and service principals | A new agent or application is in the tenant, whether anyone requested it or not |
| Permissions granted | The agent can now reach more data or APIs than before |
| Admin consent granted | One approval grants the agent its permissions across the whole tenant, either on behalf of every user or directly to the agent |
| Credentials added | Whoever holds the new secret or certificate can sign in as the agent |
| Owners added or removed | Someone new can change the agent’s configuration and credentials, or nobody is left who can manage it |
For agents built with Entra Agent ID, credentials sit on the blueprint, not on each agent. Anyone holding a blueprint’s credential can get tokens for every agent created from it.
Monitoring the identity itself is only part of the picture. An agent that can manage users or groups can also make unwanted changes to those objects. Those changes need to be found and reviewed too.
The Answers Are Spread Across Many Object Types
A simple question like “which agents got new credentials this month?” is hard to answer. An agent can run as an agent identity, an agent’s user account, a service principal, or a regular user account, and its credentials can sit on a blueprint or an app registration. Each is a different object type, and each change is recorded as a different activity.
To answer the question in Entra ID’s audit log, an administrator has to know which object types and activities are involved, then filter each one. The person on call during an incident may not know how the agent was configured.
Ask What Changed With Your Agent Identities
In Cayosoft Guardian, an investigation simply starts with a question, Ask what you want to know and Guardian surfaces the matching changes from its change history.
For example:
- “Show me agent identities created this week”
- “Show owners added to agents”
- “Show permissions granted to applications this month”
- “Show credentials added to agents”
See it in Action:
The administrator on call doesn’t need to know in advance which object holds the credential or exactly how the agent was configured. They can start with the question they need answered and investigate from there.
Roll Back the Agent Changes You Didn't Approve
Finding the change is only the first step. Agent identities, blueprints, and app registrations are all Entra ID objects, and Cayosoft Guardian lets administrators roll back changes to them in one click, whether that’s a new credential, a granted permission, or a removed owner.
The same goes for changes the agent made. If it added someone to the wrong group or edited the wrong user, Guardian can reverse the affected object or attributes without scripts or a full restore.
From Investigation to Rollback
AI agents add identities to the directory in many forms. With Cayosoft Guardian, identity teams can use AI agent identity monitoring to ask what changed in plain language, review the answer alongside the rest of their hybrid identity change history, and reverse changes that shouldn’t have happened.
See For Yourself with Cayosoft Guardian
AI is expanding the identity landscape, and identity resilience has to keep pace. Cayosoft Guardian gives identity teams the capabilities to manage, monitor, investigate, roll back, and recover from identity changes across their Microsoft environments, now extending that resilience to AI agent identities.
Learn more about Cayosoft Guardian and request a demo to see how it helps build identity resilience across your Microsoft environment.
FAQs
AI agent identity monitoring gives identity teams visibility into the identities AI agents use and the changes made to them over time. That includes changes to permissions, credentials, owners, and other configurations that can affect what an agent can access or do.
Teams should watch for changes that affect an agent’s access or control, including new agent identities and service principals, permissions or admin consent being granted, credentials being added, and owners being added or removed. It’s also important to monitor the changes agents make to other objects, such as users and groups.
AI agents in Entra ID use one of two models. Agents built with Microsoft Entra Agent ID get dedicated objects: an agent identity, an optional agent’s user account, and an agent identity blueprint that holds the credentials for every agent created from it. Other agents run on ordinary identities, such as an app registration and its service principal, or a regular user account. Investigating agent changes means covering both.
Yes. Cayosoft Guardian can roll back changes to supported Entra ID objects associated with AI agents, including creation of new identities, granting role assignments, API permissions, and owner assignments. It can also reverse changes an agent makes to other supported objects, such as users or groups.
An AI agent’s identity determines what it can access and what actions it can take. Changes to permissions, credentials, ownership, or configuration can therefore change the agent’s capabilities or who controls it. Monitoring those changes gives identity teams a way to identify and investigate activity that wasn’t expected.