What’s new in Azure Language in Foundry Tools

Every day, your applications handle numerous emails, transcripts, and documents that contain private information—such as names, contact numbers, and health records. Manually spotting and securing this data can be overwhelming. This is where the latest advancements in PII (Personally Identifiable Information) detection within Azure Language in Foundry Tools come into play. They can assist you in automating privacy protection for text, conversations, and healthcare contexts.

At Microsoft Build, we are excited to unveil new and enhanced features in Azure Language in Foundry Tools aimed at empowering developers to manage sensitive information across text, conversations, documentation, and healthcare scenarios.

Here’s a quick overview of these updates:

• The Text PII API is now generally available (GA) and includes extra redaction options and improvements in detection behaviour.
• New preview releases are available for both Text PII and Conversational PII.
• We’ve created new Microsoft Foundry playgrounds to help teams explore and evaluate these capabilities prior to API integration.
• Enhanced support for a broader variety of text and conversational scenarios has been introduced.

With Azure Language in Foundry Tools, businesses can effectively identify and manage sensitive information within textual data, transcripts, and documents while also analysing healthcare-related texts. This is particularly useful for tasks such as privacy workflows, data processing, and integrating with downstream applications.

We’re also launching several Microsoft Foundry playground experiences designed to help teams experiment with language capabilities, review output, and determine how the APIs can be integrated into their workflows before they go live.

These playgrounds offer a great environment for hands-on testing and evaluation.

Image 1: Screenshot of Text Analytics for Health Playground in use..

The GA Text PII API comes packed with updates, including new supported entity types and extra configuration options for developers to tailor how sensitive data is processed.

Some new customer customisation features include:

• Anonymization: The `syntheticReplacement` redaction policy allows you to mask detected PII entities with synthetic replacements. For instance, “John Doe received a call from 424-878-9193” can be modified to “Sam Johnson received a call from 401-255-6901.”
• Optional type validation controls: You can disable entity type validation for cases where operational efficiency outweighs the need for strict data integrity checks.
• Confidence threshold configuration: Set a minimum confidence score threshold to control the output based on the detection confidence level.
• Exclude specific values from PII output
• Utilise entity synonyms for tailored PII detection

These features offer greater flexibility in fitting PII processing to specific application needs.

Text PII (Preview)

The latest preview introduces support for more entity categories and extends coverage across various scenarios. New entities now include Passwords, PIN codes, Zip codes, and Airport codes. For a complete list, please check our public documentation.

Conversational PII (Preview)

For situations involving transcripts and dialogues, our Conversational PII redaction service is fine-tuned for spoken-language patterns, including interruptions, filler words, speaker turns, and incomplete sentences, which are common in everyday conversations and transcripts. The preview updates enhance support for these spoken-language nuances.

Improvements include:

• A new Conversational PII model (2026-04-15-preview)
• A new Conversational PII API (2026-11-15-preview) that builds on the existing model capabilities

Additional supported entities encompass GitHub account identifiers, credit card expiry dates, and zip codes. For a full list, visit our public documentation.

These preview enhancements broaden the range of inputs and scenarios supported for conversational data processing.

• Explore language playgrounds in Microsoft Foundry
• Experiment with the latest GA and preview APIs for Text PII and Conversational PII
• Check the documentation for detailed entity lists, versioning, and guidance on usage

Please note that all preview features are subject to change. Outputs generated by AI systems should be reviewed and validated for your particular requirements.

Begin building applications that prioritise privacy today—dive into the playgrounds, test out the APIs, and share your thoughts in the comments below.