All research
In-house evaluation, 2024

Detection rate against Microsoft Presidio

Our first published comparison, from November 2024, and the one that comes with a ten-page methodology PDF. Superseded in scope by the two full-dataset runs above, and kept here because its numbers still stand.

Run by Hexagone AI, November 2024. Public dataset, reproducible.

Presidio is the open-source reference for PII detection, and the honest baseline for anyone evaluating this category. We ran both systems over 5,000 samples of the public ai4privacy/pii-masking-200k dataset, in French, English, German and Italian.

Privacy score: the percentage of ground-truth PII entities successfully detected and masked. Higher is better.

The second experiment matters more than the first. Real documents contain typos, abbreviations and paraphrases. We regenerated every ground-truth entity in four distorted forms to see which system holds up.

Overall PII detection rate, Presidio against Hexagone AI
ExperimentMicrosoft PresidioHexagone AI
BaselineStandard dataset text.75.30%96.95%
AugmentedTypos, abbreviations and paraphrases, closer to real documents.65.89%93.34%
Detection rate by language
LanguageBaselineAugmented
PresidioHexagone AIPresidioHexagone AI
French74.74%97.97%65.76%94.42%
English72.77%94.66%64.10%91.14%
German76.58%96.30%68.03%92.93%
Italian76.15%97.39%64.65%93.83%

Consistency, not just average

A high average hides bad documents. On the baseline set, Hexagone AI has a standard deviation of 9.12% against 20.53% for Presidio. In practice that is the difference between a tool you can leave running on a folder and one that needs a human to check every file.

The gap widens under real conditions

Between the clean set and the messy one, Hexagone AI drops 3.73 points. Presidio drops 12.50. The advantage goes from 28.76% to 41.65%, which is the number that matters, because your documents look like the messy set.

Read the full report, methodology included

PDF, 10 pages. Datasets, prompts, distributions and per-language breakdowns. No form, no email.

Download the PDF

Run it on your own documents.

A benchmark is someone else's data. One week free, no credit card, and the files never leave your machine, so there is nothing to risk in checking for yourself.