AI Classify finds sensitive data that rule-based classification misses. It samples and examines actual data from your database, not just column names and metadata. This makes it particularly effective for columns with non-informative names where rules alone can't help.

How to use AI Classify

AI Classify can be turned on in the rganonymize CLI by entering --scan-type Ai or --scan-type RulesAndAi when running classify.

The first option (data) performs identification and classification of sensitive information using AI Classify exclusively, while the second option (RulesAndAi) combines AI Classify and the approach based on classification rules and the structure of the database.

The full command-line options for rganonymize classify are documented here. The output is a classification file in the usual format, ready for use in the rganonymize mask command.

GUI

AI Classify can be turned on in the settings page. Once enabled, AI Classify will perform a data scan on all newly created database models.

How does AI Classify work

AI Classify results take precedence over built-in and custom classification rules. Your rules are still applied, but where AI Classify identifies a column differently, its classification wins.

You can override AI Classify's decisions:

- In the GUI: change the dataset from the dropdown for any column. Your choice always takes precedence.
- In the CLI: use individual overrides in your classification JSON to make changes that persist across runs.

System Requirements

When using the Test Data Manager GUI there are no prerequisites. 

If using the command lines, there is one pre-requisite: please install the Microsoft Visual C++ 2019 Redistributable from: https://aka.ms/vs/17/release/vc_redist.x64.exe. 

Troubleshooting AI Classify

Failed to load ML model

If you see an error message Failed to load ML model, this usually means you need to install the Microsoft Visual C++ redistributable. Please follow the instructions above under system requirements.

DXGI Factory Message 

When you use AI Classify with data scanning you may see a message Failed to create DXGI factory

This message can be ignored. AI Classify will still complete successfully without any issues. 

Note

• The machine learning model is hosted on your machine so no information from your database will ever leave your environment.
• The machine learning model will not be trained on your data during use.
• There is no guarantee that AI Classify data scanning or structure scanning will find all of the sensitive data within a database. False positives and false negatives are possible.