Error when creating agent evaluation from Foundry UI

Ipsita Dutta 0 Reputation points
2026-07-14T07:09:44.2433333+00:00

User's image

I’m getting the attached error while creating an agent evaluation in Microsoft Foundry. I’ve checked the dataset, and it looks correct from my side. Could you please help me understand what might be causing this issue or verify if I’m missing any permissions?

Foundry Tools
Foundry Tools

Formerly known as Azure AI Services or Azure Cognitive Services is a unified collection of prebuilt AI capabilities within the Microsoft Foundry platform


1 answer

Sort by: Most helpful
  1. Megha Ramakrishnan 330 Reputation points
    2026-07-14T11:12:27.7133333+00:00

    Hi @Ipsita Dutta ,

    Could you please check the below storage account and access permissions.

    Storage Account:

    1. Azure AI Foundry must read your uploaded evaluation dataset from, and write logs to, the Workspace's default Storage Account.
    2. Open the Azure Portal and navigate to the Storage Account linked to your AI Foundry Workspace/Project.
    3. In the left menu under Security + networking, click Networking. If your firewall is set to 'Selected networks', scroll down to Exceptions and check the box for: 'Allow Azure services on the trusted services list to access this storage account'.
    4. Save the changes and wait 2–3 minutes before retrying the evaluation.

    Missing RBAC on the Project's Managed Identity:

    When evaluating tool calls, the platform uses the system's Managed Identity to read and parse the dataset—not your personal user account.

    1. Find the Managed Identity name associated with your Azure AI Foundry Project.
    2. Navigate to your Storage Account -> Access Control (IAM) in the Azure Portal.
    3. Click Add -> Add role assignment. Select 'Storage Blob Data Contributor'.
    4. Assign it to your Project's Managed Identity.

    Please 'Upvote'(Thumbs-up) and 'Accept' as answer if the reply was helpful. This will be benefitting other community members who face the same issue.

    Thank you!

    Was this answer helpful?

    0 comments No comments

Your answer

Answers can be marked as 'Accepted' by the question author and 'Recommended' by moderators, which helps users know the answer solved the author's problem.