Protegrity-PII model detects personal data in text with 0.915 typed recall

Key points

• Protegrity-PII is a new PII decision model for both structured and unstructured text, per an arXiv preprint submitted 21 October 2026.
• Reported typed recall of 0.915 and precision of 0.907 on a new benchmark called PII-SCOPE.
• Runs at 0.49 seconds per million characters, or $0.0005 per million characters, on a single RTX 6000 with batched inference.
• Authors credit a broad entity type taxonomy for stronger out-of-distribution generalization than the compared models, according to the abstract.

Why it matters for marketers: Content teams handling customer text could use this as a candidate tool for redacting personal data, though the results are self-reported and unreviewed.