Content Filters Not Working in Azure openAI Service

Yovel Cohen 25

I have created a Content Filter which has all of the annotations turned Off.
The "Annotate and block" is turned off completely.

Yet some prompts (with images) that I try on my endpoints (O1) are failing:
{'code': 'ResponsibleAIPolicyViolation', 'content_filter_results': {'sexual': {'filtered': False, 'severity': 'safe'}, 'violence': {'filtered': True, 'severity': 'high'}, 'hate': {'filtered': False, 'severity': 'safe'}, 'self_harm': {'filtered': False, 'severity': 'safe'}}}

Why is Azure keep filtering even if I turned off the filters?

santoshkc 12,265 Reputation points Microsoft Vendor

2025-02-13T11:12:08.2666667+00:00

Hi @Yovel Cohen,

Thank you for reaching out to Microsoft Q&A forum!
Azure OpenAI Service enforces content filtering based on Microsoft's Responsible AI policies, even if you've turned off all annotations and set "Annotate and block" to off. Some prompts, especially those involving images, may still get blocked if they trigger high-severity content categories like violence. This is a built-in safeguard that ensures compliance with responsible AI standards.

For more details, you can check out Azure OpenAI Content Filtering, which explains how the system works and what gets filtered. Additionally, the Responsible AI Practices for Azure OpenAI document covers Microsoft's broader AI policies. If you believe your prompt is being wrongly flagged, you can try rephrasing it.

I hope you understand. Thank you.

1 answer

Martin Brandl 395 Reputation points MVP

2025-02-13T10:49:57.08+00:00
Azure OpenAI still enforces baseline responsible AI policies, even if you disable content filters on your endpoint. This means that some categories, particularly high-severity violence, will still trigger filtering at the platform level, regardless of your settings.

Your error message indicates:

'violence': {'filtered': True, 'severity': 'high'}

This means the system flagged the response as high-severity violence, which is automatically blocked by Azure's internal enforcement.
Please sign in to rate this answer.
Yovel Cohen 25 Reputation points

2025-02-13T13:47:30.6666667+00:00

Thanks for the answer :)

Given the fact I asked it to summarize a scene from a 90s family sitcom (it worked without the images), I'm having a bit of hard time getting the reasoning behind the block...

santoshkc 12,265 Reputation points Microsoft Vendor

2025-02-13T14:20:30.6066667+00:00

Hi @Yovel Cohen,

While you’ve disabled content filters on your endpoint, Azure still applies a baseline Responsible AI policy that may sometimes be overly cautious. It’s possible that something in the image metadata or visual content triggered the high severity.

I hope you understand! Thank you.
Sign in to comment

Use comments to ask for clarification, additional information, or improvements to the question.

Share via

Content Filters Not Working in Azure openAI Service

1 answer

Your answer