Test user access to an AI agent

Run a manual test to verify that only the users you want to access the AI agent can do so.

Before you begin

Role required: sn_aia.admin and either admin or at least one role required by the ACL of the AI agent being testing and each of the ACLs of its downstream components

About this task

You can establish the security settings for an AI agent in the guided setup to establish which users can access it. See Define security controls for instructions on how to change the user access settings. When you select Save and continue on that step of the guided setup, an ACL is created that establishes limitations on which users can access the AI agent.

Once you have created these ACLs, you can verify that they work as intended by using the Test access test type of a manual test of an AI agent.

To see instructions for performing manual tests to evaluate performance, see Test performance manually. For more information about automated tests, see Evaluate an AI agent.

Procedure

1. Navigate to All > AI Agent Studio > Testing.
2. Select Start manual test.
3. In the Choose a test type drop-down menu, select Test access.

   If you want to test the basics of how an AI agent works, select AI agent or worfklow. The full instructions for that test type can be found in Test performance manually.

4. Search for or select the name of the AI agent you want to test.
5. Select an invoking user.

   The invoking user can be the user that triggers the AI agent or it can be the invoking user of an upstream component, such as an agentic workflow. For more information about how the invoking user works, see Security for AI agents.

   When you select an invoking user, the user roles are populated in the Invoking user roles field. The field is read-only. If you want to change a user's roles, you must change the user's User record. See Assign a role to a user.

6. Select Test access.

   

Result

The results of the access test are opened in the Access Analyzer in a new browser tab.

Access Analyzer identifies all the ACL calls made in the execution of the AI agent, including its tools, if any are present. You can see the results of each ACL to verify that each one has the correct user access defined.

What to do next

If the results are different than what you expect or want, you can redefine the security controls of the AI agent.

You can also test an AI agent's performance with either a manual test or automated evaluations. See Test performance manually or Evaluate an AI agent.