liveBook · Manning

Biometric categorization

As explained in AI Governance

Biometric categorization is the classification of people into categories using physical or behavioral biometric data.

Biometric categorization uses physical or behavioral biometric data to classify people. Under the EU AI Act, categorizing people by sensitive attributes such as race, political opinion, union membership, religion, or sexual orientation is prohibited. Using the same data to audit a training dataset for demographic fairness is permitted. section 7.2.2

Assess the purpose and use of biometric classification:

  • Inferring ethnicity from facial images to filter job applicants is an example of prohibited categorization. section 7.2.2
  • Gender is not named in the banned list. Inferring it for targeted advertising is not inherently illegal, but risk rises if it acts as a proxy for a protected class. section 7.2.2
  • Disclose biometric categorization to affected people before processing begins. Analysis of voice pitch or pace is in scope; a keyword classifier flagging “angry” is not. section 7.2.5

How AI Governance covers it

How deep it goes
Defines covered data, lists banned attributes, and explains an audit exception, a gender-related proxy risk, and disclosure examples.
Examples
Contrasts ethnicity-based applicant filtering with dataset auditing, and voice-feature analysis with keyword-based anger detection.
What you can do after
Evaluate whether a biometric classification is prohibited or requires disclosure by checking its data, purpose, and use.
How it connects
Connects biometric categorization to sensitive attributes, fairness auditing, and emotion recognition.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of AI Governance. The section links point to the parts of the book it draws on.