AI Agent from Copilot Studio Unable to Fetch Images from Document – Need Secure Resolution

(1) Share

Report

Posted on by NM-14101115-0

Hello Microsoft Community,

I have created an AI agent in Microsoft Copilot Studio, and the agent’s knowledge source is a Word document that contains both text and images. When I test the agent in Microsoft Teams, it successfully retrieves the text responses but fails to display the images that are embedded within the document.

I understand that one possible workaround is to upload the images to SharePoint or an S3 bucket and reference them via direct links. However, making these image links public raises security and compliance concerns for our organization.

Could you please suggest a secure and compliant approach to enable the Copilot agent to access and display images from internal documents (e.g., SharePoint-hosted files) without exposing them publicly?

Any official guidance or configuration recommendations from Microsoft would be greatly appreciated.

Thank you,

Nataraj M

Categories:

Building Copilot Studio chatbots in Microsoft Teams

I have the same question (0)

All responses (3)

Answers (0)

Michael E. Gernaey 53,960 Moderator on at

Like (0)

Report
Copy link

Link copied!

Hi @NM-14101115-0

If I can ask, is the agent public? I ask because the images in blob storage do not have to be "public" they can be private and use SAS tokens in the agent to load them.

So when you say Public, do you in your organization or to external people?

Ok, setting that aside, you would need to explain your Agent and share the steps you are taking and the nodes in your (trigger? tool? Topic?) etc so we can see why its not getting the images.

Was this reply helpful? Yes No
Suggested answer

Romain The Low-Code... 2,562 Super User 2026 Season 1 on at

Like (0)

Report
Copy link

Link copied!

hello :)

did you try to switch the model in the setting to go with GPT 5 ? old GPT have some difficulties with image.

if you have under 1 000 document in the target sharepoint folder i will recommand this one for image : https://learn.microsoft.com/en-us/microsoft-copilot-studio/knowledge-unstructured-data

if you want the way of reading image 200% accurate with a lot of AI skill, it's using an Azure AI Search Index with dedicated skill for image inside document (like document intelligence capacity etc) but it will add few cost since the Free subscription is soft lock to block connexion with copilot studio.

but my first bet would be : try the link with unstructured data :) (if you have more thant 1000 doc, just split by folder, you could have 25 sharepoint connector inside an agent)

Was this reply helpful? Yes No
Michael E. Gernaey 53,960 Moderator on at

Like (0)

Report
Copy link

Link copied!

In this case the issue does not have anything to do with AI, or the model, its an intentional and necessary security feature that is intentional.

The agent automatically skips these images, so there is no way to "turn" it on or swap the model and it just works.

Its a platform level security thing they did on purpose. Don't ask me why it just is.

So the answer is no, there isn't a way to do what you want based on what you explained. Not how you want atleast

Was this reply helpful? Yes No