Efficient model finds personal data with high accuracy and low cost
Broad and Efficient PII Detection with Unified Labels
Computation and LanguageCryptography and Security
Summary
Finding personal information like names, IP addresses, or IDs in text is tricky because sources differ and labeling can be inconsistent. The authors addressed this by combining many datasets into one big, consistent training set with clear rules on what counts as personal data. They trained a model that not only finds more sensitive information, including cases in different languages and tricky formats, but also does so using much less computing power. This makes it practical to detect personal information quickly and reliably in large amounts of data.
What this means in practice
- •For data privacy teams: Improve automated detection of personal information across varied text sources with fewer false alarms and lower computation needs.
- •For cloud service providers: Deploy faster and more resource-efficient PII detection to monitor and protect customer data at scale with high accuracy.
Authors
Andy Hsu, Konstantin Berlin
Abstract
Personally Identifiable Information (PII) detection models must efficiently find identifiers across diverse contexts while maintaining a low false-positive rate (FPR) on high-volume traffic. Recent detectors increasingly rely on complex architectures that raise deployment cost, yet many are fine-tuned on only a few synthetic datasets that limit their coverage. Combining public datasets is challenging because their taxonomies differ, and even after categories are mapped to a common schema, labels can disagree on what counts as a detection, producing inconsistent training and evaluation signals. To address these problems, we present a practical 141M-parameter encoder-based detector trained on 1.12 million examples from 77 public datasets, relabeled under a detailed definition of broad contextual detection for 12 structured PII categories. Evaluated on the AI4Privacy and Nemotron-PII benchmarks at a low-FPR operating point, our model achieves the highest recall against larger baselines in 11 of 14 comparisons under our labels and 9 of 14 under the original labels. Its largest recall gains include useful additional coverage, such as multilingual disclosures and IP addresses embedded in URLs. It achieves this with 2-10x higher throughput and 94-97% less peak GPU memory than the baselines, delivering leading detection performance at substantially lower cost.