Azure OpenAI API - content filtering problems

Question

Azure OpenAI API - content filtering problems

Andy Bell 40

We have adjusted our content filtering as much as we can so it only blocks calls which register as high risk. 2 days ago we ran a prompt which was blocked by content moderation for one area. We read through the prompt content, and there is no evidence of what was considered high risk - we believe this is a false positive. Even so, we redacted anything that might be misconstrued (the redacted parts were a real stretch) but the prompt was still blocked. In addition to this false positive, we are using the LLM in an area where it is likely the prompts we run will feature discussions with content in the content moderation area, but these are expected input that we do not want to be blocked and are not a concern to us. Can we:

Make sure this blocked prompt is run ASAP?
Have our content moderation requirement removed?

SRILAKSHMI C 19,005 Reputation points Microsoft External Staff Moderator

2026-04-30T15:13:01.1233333+00:00
Hello @Andy Bell

Thank you for reaching out. I understand how frustrating this can be, especially when legitimate business content is incorrectly blocked by content filtering.

Based on your description, you are dealing with two related concerns:

A prompt that appears to have been blocked due to a false positive

A need for more flexible content filtering for your use case

Why This Happens

Azure OpenAI applies built-in content filtering to all prompts and completions across categories such as violence, hate, sexual content, and self-harm.

Even when you configure a custom content filter to block only High severity content, any prompt classified as High severity in a filtered category will still be blocked automatically.

In some cases, legitimate or contextually appropriate business content may be incorrectly classified. This is considered a false positive.

1. How to Investigate and Unblock the Prompt

Suspected false positives can be reported to cscraireport@microsoft.com

This helps the product team review and improve classification accuracy.

2. How to Relax Content Filtering

You can further tune your existing content filter in Azure AI Foundry:

Navigate to: Azure AI Foundry → Project → Guardrails + Controls → Content Filters

Edit or create a filter

Configure the input thresholds so that only High severity content is blocked

Assign the updated filter to your deployment under: Models + Endpoints → Deployment → Edit

This allows low- and medium-severity content to pass through while still blocking only high-risk content.

3. Can Content Filtering Be Completely Removed?

For standard Azure OpenAI deployments, core content filtering cannot be fully disabled by default.

However, eligible managed customers may apply for Modified Content Filtering through Microsoft's Limited Access Review process. Depending on approval, this may allow:

Annotate only mode (content is flagged but not blocked)

Reduced blocking for selected categories

In some approved scenarios, significantly less restrictive filtering

Approval is based on your use case, risk controls, and compliance requirements.

Next Steps

Submit the blocked prompt and content_filter_result for review

Report the false positive to cscraireport@microsoft.com

Continue using your current custom filter with High-only blocking

If your use case requires less restrictive filtering, apply for Modified Content Filtering

Please refer this

Configure and manage content filters in Azure OpenAI Studio: https://learn.microsoft.com/azure/foundry-classic/openai/how-to/content-filters

Content filtering concepts & risk categories: https://learn.microsoft.com/azure/foundry-classic/foundry-models/concepts/content-filter

Limited Access Review: Modified Content Filters (for disabling filters): https://ncv.microsoft.com/uEfCgnITdR Report problematic or false-positive content: cscraireport@microsoft.com

Azure OpenAI Acceptable Use Policy: https://learn.microsoft.com/legal/ai-code-of-conduct

I Hope this helps. Do let me know if you have any further queries.

Thank you!
SRILAKSHMI C 19,005 Reputation points Microsoft External Staff Moderator

2026-05-01T11:31:00.67+00:00

Hi @Andy Bell

Did you get any chance to review the above response. Do let me know if you have any further queries.

Thank you!

1 answer

Your answer

SRILAKSHMI C 19,005 Reputation points Microsoft External Staff Moderator

2026-05-01T11:31:00.67+00:00

Hi @Andy Bell

Did you get any chance to review the above response. Do let me know if you have any further queries.

Thank you!

Answer 1

Andy Bell 40

Hi

We applied for Azure OpenAI Modified Guardrails. Our request was turned down. The stated reason was: "Application Denial Reason (Optional): Unmanaged". The message also said "We are unable to accept applications to obtain managed customer status at this time."

This is a big problem for us, as our startup (https://trellis.scot) intrinsically needs to process sensitive data. We've been building for the last year, and we are getting promising traction. This is a serious blocker for us.

Your thoughts would be much appreciated!

Best

Andy

0 comments

Share via

Azure OpenAI API - content filtering problems

1 answer

Your answer