Thanks for the notification, i have no direct answer to help you, i will try to sumarize every things i could suggest to help :) (i asked copilot to help me to rewrite because we had a lot of exchange in private together about this problem and i m not native english speaking, so "disclaimer : the next part as been rewrite by copilot to correct my poor english")
Summary of what i try to help :) ->>
I’ve seen a very similar issue recently. Microsoft has made major internal changes to the content filtering system used by Copilot Studio (Responsible AI). These modifications were not widely announced, but they started to appear in the past few days. This may explain why multi-agent scenarios suddenly started failing with “ContentFiltered”.
Microsoft is also updating the multi-agent feature itself. You may have noticed that sometimes the “Add agent” window turns completely white. A temporary workaround for that is to add “preview” at the end of the Copilot Studio URL (for example /?preview=true) and the page usually loads correctly again.
Here are a few resources and things that helped in my case.
Official Microsoft reference:
https://learn.microsoft.com/en-us/troubleshoot/power-platform/copilot-studio/generative-answers/agent-response-filtered-by-responsible-ai
Be careful with Azure Application Insights as mentioned in the article. It is useful for diagnostics, but it can generate costs very quickly once enabled.
Community article worth checking:
https://iiu.dk/2025/09/18/copilot-studio-contentfiltered
This one gives some background and suggestions on how to handle the “Status Cancelled – ContentFiltered” error between agents.
From my own experience a few months ago, I had exactly the same behavior. Everything worked fine and suddenly some messages were filtered. The following actions solved it in different cases:
-
Remove all text from the description and global prompt, then test again. If it starts working, re-add the content line by line until you find which sentence or word triggers the filter. In my case one hidden phrase was enough to block everything.
-
Recreate the agent with a different default language. I don’t know why, but switching the default language (for example English instead of French) completely removed the filtering. It seems some moderation layers are language-specific.
-
Change the default moderation configuration of the agent. In one case, simply adjusting the Responsible AI / moderation setting solved the problem. The filtering logic sometimes becomes stricter depending on that setup.
Since Microsoft is currently updating both the Responsible AI filters and the multi-agent orchestration layer, it’s likely that some temporary inconsistencies are appearing (sometimes working, sometimes not). I would suggest testing with a simplified prompt, a different default language, and an adjusted moderation level. If the problem remains, please share the minimal reproducible steps, as it could help the product team identify if this is a regression in the new content filtering mechanism.
Hope this helps some one :) i don't have the "exact answer" but everything here help me sometime when having same problem :)
