Personal data you can finally work with.

The dataset catalogue.

Nine releases and counting, across text and audio, from compact test sets to the largest open collections of artificial personal data available.

Does training on it work?

Yes, and here is one number. A model trained on these datasets recognises personal data at 0.92 F1 on a held-out test set; the same architecture off the shelf manages 0.30. Same test, same scoring. The training data is the difference.

Internal evaluation, July 2026. Methodology notes and reproduction scripts ship with the public benchmark suite, in preparation.

Used in research worldwide.

• Google DeepMind · University of Maryland · Oregon State Privacy Backdoors: Enhancing Membership Inference through Poisoning uses PII-Masking-200k
• Indiana University Bloomington · JD Cloud · Stanford The Janus Interface: How Fine-Tuning in LLMs Amplifies Privacy Risks uses PII-Masking-300k
• Mitsubishi Electric Research Laboratories · Penn State Forget to Flourish: Leveraging Machine-Unlearning on Pretrained Models uses PII-Masking-200k
• Northeastern University Rescriber: Smaller-LLM-Powered User-Led Data Minimization uses PII-Masking-200k

137 published studies used these datasets. The full list is in the Gazette.