Responsible AI restriction

Question

Responsible AI restriction

Kalla, Jahnavi 60

Hi Team,

I have developed an agent that provides internal definitions for acronyms used within our organization. However, I encountered an issue while testing it.

When I query the agent in lowercase, it returns the error: “The content was filtered due to Responsible AI restrictions. Error Code: ContentFiltered.”

Interestingly, when I ask the same question using uppercase the agent responds correctly with the abbreviation and definition.

I am unsure why this inconsistency is occurring and would appreciate your guidance on the root cause and how to resolve it.

Thank you for your support.

Regards,

Ashlesha Pawar (Tata Consultancy Services Limited) 320 Reputation points Microsoft External Staff

2026-06-19T09:51:03.9833333+00:00
Hi,

Thank you for reporting this issue.

This behavior is likely caused by the Responsible AI (RAI) content filtering layer treating lowercase input differently from uppercase acronyms, leading to a false positive classification.

In some cases, moderation and intent detection pipelines normalize or tokenize input differently based on casing, which can result in inconsistent scoring for identical acronyms (e.g., “otd” vs “OTD”).

To mitigate this, you can try:

Normalizing all user inputs to uppercase before processing

Explicitly defining acronyms in system instructions (e.g., “OTD = On-Time Delivery”)

Ensuring your knowledge base contains clear acronym mappings

Reviewing moderation or safety configuration settings if available

Once applied, please retest with both lowercase and uppercase inputs to confirm consistent behavior.

Let me know if you need help reviewing your setup.

1 answer

Your answer

Ashlesha Pawar (Tata Consultancy Services Limited) 320 Reputation points Microsoft External Staff

2026-06-19T09:51:03.9833333+00:00

Hi,

Thank you for reporting this issue.

This behavior is likely caused by the Responsible AI (RAI) content filtering layer treating lowercase input differently from uppercase acronyms, leading to a false positive classification.

In some cases, moderation and intent detection pipelines normalize or tokenize input differently based on casing, which can result in inconsistent scoring for identical acronyms (e.g., “otd” vs “OTD”).

To mitigate this, you can try:

Normalizing all user inputs to uppercase before processing

Explicitly defining acronyms in system instructions (e.g., “OTD = On-Time Delivery”)

Ensuring your knowledge base contains clear acronym mappings

Reviewing moderation or safety configuration settings if available

Once applied, please retest with both lowercase and uppercase inputs to confirm consistent behavior.

Let me know if you need help reviewing your setup.

Answer 1

Dear @Kalla, Jahnavi,

Thank you for posting your question in the Microsoft Q&A forum.

Please understand that our forum is a public platform, and we will modify your question to cover your organization domain name in the description. Please notice to hide these personal or organization information next time you post error or some information to protect personal data.

Based on your description, this behavior is caused by Responsible AI (RAI) content filtering, not an issue with your agent logic.

In your case, the lowercase input “otd” is likely being misinterpreted by the moderation system as a sensitive or ambiguous token, triggering a false positive. Similar cases have been reported where valid business acronyms (e.g., “rtp”) were incorrectly classified under restricted categories.

When using uppercase “OTD”, the system is more likely to recognize it as an acronym, so the response is allowed.

Here are some suggestions you can refer to:

Normalize user input Convert queries to uppercase before processing (e.g., “otd” > “OTD”).
Define acronyms explicitly Add definitions (e.g., “OTD = On-Time Delivery”) in your agent instructions to reduce ambiguity.
Review content moderation settings If applicable, adjust moderation sensitivity to reduce false positives.
Check knowledge sources Ensure your grounded data does not contain patterns that could trigger filtering.

For detailed information, you can refer to this article Resolve responsible AI content filter errors in Copilot Studio

I hope my information meet your expectations, please feel free to reach out if you have any other questions.

If the answer is helpful, please click "Accept Answer" and kindly upvote it. If you have extra questions about this answer, please click "Comment".

Note: Please follow the steps in our documentation to enable e-mail notifications if you want to receive the related email notification for this thread.

Share via

Responsible AI restriction

1 answer

Your answer