Azure OpenAI output filter blocks original company data analysis without usable diagnostics

Question

An Azure OpenAI Responses API deployment is repeatedly blocking an original internal report generated only from company-owned sales and inventory data plus company-authored operating instructions.

The prompt filters pass. The completion returns HTTP 200 with status=incomplete and reason=content_filter, and identifies protected_material_text. A second attempt explicitly required original phrasing and prohibited quotations or reproduction of source phrasing, but it was blocked the same way.

No external source was supplied. Azure returns no matching passage or source, so we cannot determine what triggered the filter or how to remediate it.

Can Microsoft review this as a possible false positive and provide a supported mitigation? We have retained the Azure request IDs and response IDs and can provide them privately to Microsoft support.

Answer 1

Based on the behavior you've described, this appears more like a service-side content filtering decision than an application or prompt construction issue.

Since the request completes with HTTP 200, but the response status is incomplete with reason=content_filter and protected_material_text, it indicates that the Azure OpenAI safety system is blocking the generated completion rather than rejecting the request itself.

Because:

• the prompt passes validation,
• no external content or copyrighted source was supplied,
• the second attempt explicitly required original wording,
• and Azure does not return any matching passage or diagnostic explaining the trigger,

this could represent a false positive or a limitation of the current protected material detection.

I would recommend:

1. Retain the Request ID and Response ID for every affected request.
2. Open an Azure Support ticket so Microsoft can investigate the backend logs associated with those IDs.
3. Test whether the behavior is reproducible with:
   • a different Azure OpenAI model version,
     • another Azure OpenAI deployment,
       • and a simplified version of the same prompt.
       1. If the issue is reproducible across deployments while using only original company-owned data, include that information in the support request, as it will help Microsoft determine whether this is a service-side false positive.

At the moment, I am not aware of a customer-configurable setting that allows the protected_material_text filter to be selectively disabled or tuned, so this would require investigation by Microsoft if the behavior persists.