glossaryPage.heroH1

Accuracy depends on lighting, camera quality, mask-wearing, age range, and the underlying face-recognition model. NIST FRVT (Face Recognition Vendor Test) ongoing benchmarks report best-in-class commercial models at 99%+ accuracy in controlled adult-enrolment conditions, but accuracy drops materially for under-12 children (where features change rapidly), darker-skinned faces (per NIST 2019 demographic-bias study, 10-100x higher false-match rates for African-American and Asian faces), and uncontrolled school-corridor lighting. Schools deploying AI attendance should require per-demographic accuracy auditing per NIST AI Risk Management Framework, not rely on vendor-quoted aggregate accuracy.

Jurisdiction-specific. In the EU, GDPR Article 9 treats biometric data as a special-category requiring explicit consent and strong safeguards; the EU AI Act 2024 classifies face-recognition systems in education as high-risk requiring conformity assessment. In the US, the Illinois BIPA (Biometric Information Privacy Act) requires written consent and per-student opt-out; New York and several other states restrict face-recognition in K-12 schools entirely. The UK ICO has issued formal warnings and enforcement notices against schools deploying face-recognition without proper lawful basis. The EFF and ACLU have documented enforcement actions against school deployments. Schools should consult legal counsel and per-jurisdiction regulation before deployment.

Face-recognition systems exhibit documented demographic-bias: NIST 2019 FRVT demographic study confirmed 10-100x higher false-match rates for African-American, Asian, and Native American faces compared to white-male faces. Female faces also exhibit higher false-match rates than male faces. In a school context, the bias translates to disproportionate misidentification of students from specific demographic groups, generating disproportionate false absence flags and parent-alert noise. NEPC (National Education Policy Center) has published critical-perspective research on the disproportionate-impact concern. Per NIST AI Risk Management Framework and UNESCO AI 2024 framework, per-demographic accuracy auditing is a baseline requirement; vendors that cannot provide per-demographic accuracy data are flagging a governance concern.

RFID-card tap-in (low-cost, no biometric data, but susceptible to proxy attendance via card-sharing), Bluetooth-beacon or BLE-token tracking (passive, low-cost, no biometric data), QR-code self-scan via student-phone (low-cost, requires student-phone, susceptible to remote-scanning by classmate), Wi-Fi MAC-address scanning (passive, requires per-student device registration), and traditional teacher manual entry via a mobile app (lowest tech, no bias concern, costs 1-3 minutes of class time). Most schools find a hybrid of RFID and teacher manual override gives the operational benefit without the biometric-data concern.

Per UNESCO AI in Education 2024 guidance and NIST AI Risk Management Framework: (1) document the deployment justification and weigh against less-intrusive alternatives, (2) obtain explicit parent and student consent per jurisdiction-specific law, (3) require per-demographic accuracy auditing from the vendor with quarterly review, (4) provide a per-student opt-out path without academic penalty, (5) restrict data retention to the operational minimum, (6) prohibit secondary use beyond attendance (no behavioural-pattern analysis without separate consent), and (7) maintain a human-in-the-loop review for any consequential action (chronic-absence intervention) triggered by AI-flagged data. Schools without clear governance should not deploy face-recognition attendance.