Power Platform Community Forum Thread Details

Hi everyone,

I’m facing an issue while working with multiple Copilot Studio agents and would appreciate some help.

Setup:

We have two main agents:
- C1 (Parent Agent)
- C2 (Child Agent of C1)
Inside C2, there are 3 additional child agents.
Flow:
1. Conversation starts from C1
2. C1 calls C2
3. C2 performs content generation and processing
4. C2 returns the status back to C1

Issue:

When C1 calls C2, the process starts correctly.
C2 performs partial processing, but then suddenly fails.
It shows the error:

"Blocked Step – The content was filtered due to Responsible AI restrictions."

Problem:

The workflow does not complete.
Execution stops midway with this “Blocked Step” error.

What I Need Help With:

Why is this happening during agent-to-agent communication?
Is this due to:
- Content size?
- Prompt structure?
- Nested child agents?
Any best practices to avoid this issue?

I’ve also attached a screenshot of the error for reference.

Thanks in advance!

Categories:

Calling actions from Copilot Studio

Autonomous agents

Copilot Studio pre-built agents/templates

Copilot Studio skills development

@AA-06040553-0,

The error is triggered because Microsoft’s Responsible AI safeguards have identified content that may violate safety or compliance policies.

Could you clarify which moderation level you’ve configured—Low, Medium, or High?

This issue commonly occurs when your knowledge source contains wording that sounds instructional or directive, such as “Encourage all users” or “Pin your hub sites.” It can also arise when Generative AI orchestration is used together with email triggers. This combination is treated as high risk, since it may result in unreviewed content being sent externally.

Responsible AI filters are designed to block any content that could be interpreted as unsafe, sensitive, or potentially harmful.

To mitigate or resolve the issue, start by refining your knowledge source:

Remove imperative or directive language like “Encourage all users” or “Pin your hub sites.”
Avoid embedded hyperlinks, or replace them with plain text.
Keep the language neutral and descriptive rather than instructional.
If possible, disable Generative AI orchestration for email-triggered flows and instead use static responses or classic Q&A patterns.
Always include a fallback response, for example: Sorry, I’m unable to process this request at the moment. and Additionally, you can review logs in Azure Application Insights to pinpoint the specific content that triggered the Responsible AI filter.

Links