Map agent privileges to cloud resources in Identity Security Insights

A workflow for Insights

An agent's endpoint privileges are only half the picture. What it can reach through the identities and entitlements attached to it is where the real blast radius lives; an agent with a modest local footprint can hold a path to production cloud resources through the account that runs it. This workflow shows you how to map a discovered agent's privilege out to the cloud resources it could reach.

Scope of this guide

This workflow covers tracing an agent's privilege and access paths in Identity Security Insights. It does not cover connecting cloud providers or remediating the entitlements you find.

Prerequisites

Identity Security Insights is in use in your environment.
Your cloud and identity providers are connected as Insights connectors.
You have identified an agent worth tracing, from your AI Agent Security inventory or from a session.

Steps

Sign in to app.beyondtrust.io.
Select the navigation menu, then Insights > AI Agents.
On the Privileged AI Agents by Provider tile, review privilege levels across your connected platforms.
On the Risky AI Agents tile, select the agent you want to trace. The agent panel opens.
Review Agent Privileges to see the permissions granted to the agent, and spot overprovisioned access.
Review Security Graph to see which accounts can access the agent and how it connects across your environment. This is where privileged chains and unexpected exposures become visible.
Review Who Can Access This Agent for the roles, entities, and other agents that can invoke it. An agent invoked by another agent is a cascading path to privilege.
Review Tools for the callable tools the agent reaches, such as MCP servers or third-party actions.
Review Knowledge Sources for the data shaping the agent's behavior. Outdated or confidential sources affect both accuracy and exposure.
Review Entitlements to assess the agent's exposure by privilege level and type.
Review Recommendations, which are prioritized by privilege level and severity.

Verify the results

Confirm you have identified at least one agent-to-resource path that was not already known (a cloud resource the agent can reach, or an identity that can invoke it, that nobody had accounted for).

Next steps

Work the Recommendations for the agent, highest severity first.
Trace the accounts listed under Who Can Access This Agent and confirm each should be able to invoke it.
Compare the agent's cloud entitlements against what its work actually requires, and reduce the difference.