Biometric data

On this page 7

Biometric data is personal data that results from specific technical processing of a person’s physical, physiological or behavioral characteristics, such as a fingerprint, a face or a voice, and that allows or confirms the unique identification of that person. In digital identity systems, it links an account, a credential or a wallet to a living human being rather than to something the person knows or carries.

Privacy law treats this category with exceptional care because biometric traits are permanent and cannot be reissued. A leaked password can be changed in minutes, but a compromised fingerprint template stays tied to its owner for life, which is why the General Data Protection Regulation (GDPR) and laws such as the Illinois Biometric Information Privacy Act (BIPA) impose strict conditions on its collection and use.

How biometric data is captured and processed

A biometric system converts a raw sample into a compact mathematical representation called a biometric template. A sensor captures an image or signal, software extracts distinctive features, and the system stores or compares the resulting template rather than the original picture in well-designed deployments.

The legal definition follows this technical pipeline. Under Article 4(14) of the GDPR, an ordinary photograph is not automatically biometric data; it becomes biometric data only when processed through specific technical means that enable unique identification or authentication. The table below traces each processing stage and its privacy relevance.

StageOutputPrivacy relevance
CaptureRaw image, audio or signalMay reveal health or ethnicity
Feature extractionMinutiae, landmarks or embeddingsCreates identifying data
EnrollmentStored reference templateLong-lived and high-value target
ComparisonSimilarity scoreThreshold choices drive error rates
DecisionMatch or no matchMay trigger access or legal effects
Stages of the biometric processing pipeline

Two operating modes must be distinguished. Verification (one-to-one) checks whether a person matches a single claimed identity, as when a phone unlocks for its owner. Identification (one-to-many) searches a whole database for a match, which carries far greater risk because it enables covert recognition of people who never chose to participate.

Types of biometric characteristics

Biometric modalities fall into physiological traits, which describe the body, and behavioral traits, which describe how a person acts. Each modality offers a different balance of accuracy, convenience, user acceptance and exposure to spoofing.

ModalityTrait typeTypical useMain weakness
FingerprintPhysiologicalDevice unlock, border controlLatent prints can be lifted
FacePhysiologicalOnboarding, payments, surveillanceBias across demographic groups
IrisPhysiologicalHigh-security access, national IDRequires specialized cameras
VoiceBehavioral and physiologicalCall center authenticationVulnerable to synthetic speech
Keystroke dynamicsBehavioralContinuous fraud detectionLower accuracy, varies with context
Common biometric modalities compared

Accuracy is measured with two linked metrics: the false match rate (FMR) and the false non-match rate (FNMR). Lowering the decision threshold reduces rejected genuine users but admits more impostors, so every deployment makes an explicit security versus convenience trade-off that should be documented in its risk assessment.

Most modern privacy laws classify biometric data as a sensitive or special category. Under Article 9 of the GDPR, processing biometric data for the purpose of uniquely identifying a person is prohibited unless a specific exception applies, most commonly explicit consent, employment law obligations or substantial public interest defined in law.

In the United States, regulation is fragmented. BIPA requires written notice, a written release and a published retention schedule, and it grants a private right of action with statutory damages, which has produced extensive class action litigation. Texas and Washington have narrower biometric statutes, and several comprehensive state laws treat biometric data as sensitive data requiring opt-in consent.

FrameworkClassificationCore requirement
GDPRSpecial category when used for identificationArticle 9 exception plus a DPIA in most cases
Illinois BIPABiometric identifiers and informationWritten release and retention schedule
CCPA and CPRASensitive personal informationNotice and right to limit use
EU AI ActHigh-risk or prohibited usesBans most real-time remote identification in public
Key legal frameworks for biometric data

The EU AI Act adds a second layer on top of data protection law. It prohibits untargeted scraping of facial images to build recognition databases, restricts real-time remote biometric identification in publicly accessible spaces for law enforcement, and classifies many other biometric systems as high-risk AI.

Biometric data in digital identity wallets

Digital identity architectures use biometrics mainly to bind a credential to its holder. During identity proofing, a selfie is compared with the photo in a passport chip, and liveness detection confirms that a real person, not a mask or replayed video, is present.

Privacy-preserving designs keep templates on the user’s device, typically inside a secure enclave, and release only a yes or no result to the relying party. This model, used in the EUDI Wallet framework and in FIDO-based passkeys, avoids building a central biometric database and aligns with data minimization and selective disclosure principles.

  • On-device matching limits breach impact to a single device rather than millions of users.
  • Template protection under ISO/IEC 24745 makes stored templates renewable and unlinkable across systems.
  • Presentation attack detection under ISO/IEC 30107 tests resistance to photos, masks and deepfakes.

Risks and common misconceptions

The most persistent misconception is that templates are anonymous because they are not images. Research has repeatedly shown that face images can be partially reconstructed from unprotected templates, so a template remains personal data and must be secured accordingly.

RiskConsequenceSafeguard
Central database breachPermanent compromise of identitiesDecentralized or on-device storage
Function creepAccess data reused for surveillanceStrict purpose limitation
Demographic biasHigher error rates for some groupsIndependent accuracy testing
Spoofing and deepfakesImpersonation of enrolled usersCertified liveness detection
Biometric risks and appropriate safeguards

Another mistake is relying on consent where the person has no genuine choice, such as employee timekeeping with fingerprint scanners. Supervisory authorities have repeatedly found such consent invalid because of the imbalance of power, so employers should offer a non-biometric alternative.

Explained without jargon

Biometric data is information about your body or behavior. A computer uses it to recognize you, for example by your fingerprint, face or voice. It works like a key that is part of you, so you never forget it at home.

The problem is that you also cannot replace it if it is stolen. That is why the law treats it as extra sensitive. Companies must collect it only when truly needed and keep it safe. Ideally, it stays on your own phone, not on their servers.

A typical use case

A European bank lets new customers open an account entirely online. The applicant photographs a passport, the app reads the photo stored in the passport chip, and the customer records a short selfie video. A liveness check confirms a live person, and a face comparison returns a match score above the bank’s threshold.

Because this processing counts as special category data, the bank completed a data protection impact assessment, collects explicit consent, and offers a branch visit as an alternative. The selfie and the template are deleted after 30 days once onboarding is confirmed, keeping only the verification result. The bank meets its anti-money-laundering duties while keeping its biometric footprint small and defensible.