Share via

Default Content Filters Blocking Legitimate Personal Data in Real-Time GPT Bots

Abdul Rehman 20 Reputation points
2025-12-05T15:47:08.8466667+00:00

We are using Azure OpenAI in real-time GPT bots that handle personal information, such as phone numbers.

Recently, the default content filters occasionally block or modify these legitimate inputs. This causes:

  • Missing conversation transcripts in emails
  • Incorrect or default values in SMS and downstream workflows

Disruption of real-time bot interactions

Details

  • Default content filters (defaultV2) cannot be modified or disabled.
  • User prompts containing personal data trigger the filter, while system prompts are less affected.
  • Custom content filters can be created but cannot override the default filter.
  • Our workflows require handling personal data safely.

Questions / Request

  1. How can we reliably process personal information (phone numbers, emails, etc.) in real-time GPT bots without being blocked?
  2. Is there a way to request exceptions, higher PII tolerance, or guidance for enterprise subscriptions?
  3. What are Microsoft’s recommended best practices for handling sensitive data with Azure OpenAI in real-time applications?

Impact

  • Business-critical workflows are disrupted
  • Real-time bots cannot process valid customer information
  • Emails and SMS notifications sometimes lack correct conversation data

Request

We request guidance or recommendations on how to handle PII safely in real-time mode without disruption.

Azure OpenAI Service
Azure OpenAI Service
An Azure service that provides access to OpenAI’s GPT-3 models with enterprise capabilities.
Answer accepted by question author
  1. Anshika Varshney 5,365 Reputation points Microsoft External Staff Moderator
    2025-12-05T16:45:16.5733333+00:00

    Hello Abdul Rehman,

    Thank you for reaching out on the Microsoft Q&A.

    The issue you’re experiencing is caused by Azure’s default content filters, which apply medium-severity thresholds to all deployments. These defaults can sometimes block legitimate inputs especially when the text contains personal information or sensitive-looking data.

    To address this, Azure allows you to create your own custom content filters. In Azure AI Foundry, you can go to Guardrails + controls → Content filters and create a filter where you adjust the severity levels for each safety category. After creating the filter, you can attach it to the specific model deployment. This usually helps reduce unnecessary blocking in scenarios where your application deals with real user data.

    However, it’s important to note that the default system filter still applies for most customers. Unless you have special approval, you cannot fully override or disable the underlying default filter. Azure requires customers to apply for “modified content filtering” if they need deeper control, such as more relaxed filtering or annotation-only behavior.

    If your scenario involves frequent PII or sensitive inputs, the recommended approach is to first configure a custom filter and see if it resolves your issue. If not, then you should apply for modified content filtering as described in the official documentation. Without that approval, the default filter may continue blocking certain types of content.

    Reference:

    Configure content filters - Azure OpenAI | Microsoft Learn

    There is an option to configure content filters, and if you are eligible, you can apply to modify these filters as mentioned on the page.

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

    Thankyou!

    0 comments

2 additional answers

Sort by: Most helpful
Most helpful Newest Oldest
  1. Q&A Assist
    2025-12-05T15:47:24.34+00:00
    AI generated content. This question contains an answer created with AI from Q&A Assist. Learn more

    To reliably process personal information such as phone numbers and emails in real-time GPT bots using Azure OpenAI, consider the following approaches:

    1. Custom Content Filters: While default content filters cannot be modified or disabled, you can create custom content filters that are tailored to your specific use case. However, these custom filters cannot override the default filters. You may need to design your workflows to minimize the triggering of default filters by adjusting how personal data is presented in prompts.
    2. Request for Modified Content Filtering: For enterprise subscriptions, you may be able to request exceptions or higher tolerance for PII through the appropriate channels. This typically involves applying for modified content filters, which allows for more control over content filtering settings. You can submit a request via the forms provided by Microsoft for Azure OpenAI models.
    3. Best Practices for Handling Sensitive Data: Microsoft recommends following best practices for responsible AI usage, which include:
      • Conducting thorough testing to identify potential issues with content filtering.
      • Implementing iterative identification and measurement processes to understand the impacts of content filtering on your workflows.
      • Ensuring compliance with the Code of Conduct when integrating Azure OpenAI into your applications.
      • Utilizing annotation features to help manage and understand filtered content without losing critical information.

    By following these strategies, you can better manage the handling of personal data in your real-time applications while minimizing disruptions caused by default content filters.

    References:

    0 comments
  2. Abdul Rehman 20 Reputation points
    2025-12-16T08:45:14.45+00:00

    “(Removed — no longer applicable)”

    0 comments

Your answer

Answers can be marked as 'Accepted' by the question author and 'Recommended' by moderators, which helps users know the answer solved the author's problem.