Dynamics 365 Contact Center introduces large language model (LLM)‑based sensitive data redaction for human‑to‑human interactions, allowing administrators to automatically redact PCI data shared during live conversations between customers and customer service representatives.
The system uses LLMs to identify sensitive data based on conversational context, not just patterns, across both voice and digital channels. When sensitive data such as a credit card number, PIN, or account number is detected, it is replaced with a standardized redaction token (for example, CreditCardNumber:Redacted) before the data is persisted.
Redaction is applied consistently across:
Conversation transcripts
Call recordings
Logs and diagnostic data
Analytics outputs derived from conversations
Administrators can configure which PCI data types are redacted, and the system enforces these policies automatically without requiring representatives to pause recordings or change their workflows. This capability helps ensure that sensitive customer data isn't stored while preserving the non‑sensitive conversation context needed for quality assurance, analytics, and operational insights.
Business Value:
Organizations in regulated industries must prevent sensitive payment data from being stored in call recordings, transcripts, logs, or analytics artifacts. Today, customers rely on manual processes or limited rule‑based masking, which increases compliance risk and operational overhead. This feature enables customers to safely use recording, transcription, and AI analytics by ensuring payment card information (PCI) data shared during live conversations between customers and customer service representatives is automatically redacted before storage. As a result, customers can reduce PCI scope, prevent compliance escalations, and confidently adopt advanced contact center capabilities without disrupting agent workflows or customer experience.
Enabled for:
Enabled by admins/makers/marketers for end users
GA date: March CY2027
Preview date: December CY2026