Perplexity's PII-TRACE & PII-Tracer: On-Device Privacy

Perplexity has introduced a new benchmark called PII-TRACE and an on-device detector called PII-Tracer. This new technology aims to protect personally identifiable information, or PII. Here's the thing: PII-Tracer is a compact model designed to flag PII on a user's device before text is sent to cloud models. This is part of Perplexity's hybrid compute architecture, where local models handle private files. The local system can keep sensitive content private, redact information, or ask for approval before sending it to the cloud. What's interesting is that detecting PII becomes harder in long, multilingual conversations where the same identifier might appear many times. One missed mention could expose sensitive information. The PII-TRACE benchmark helps evaluate PII detectors. It includes over 13,000 synthetic conversations in 13 languages, with more than 37,000 labeled identifier mentions across nine PII types. This benchmark focuses on consistent coverage, robustness to long text, and handling multiple languages and mixed content. The bottom line is this system is designed to enhance privacy by identifying and managing sensitive data directly on your device.