Identity Verification

Impersonation is a different risk from assistance, and the measures against it are different. Conflating them produces systems that do both badly. The identity and audit questions can be compared with this implementation guide, especially when names, accounts or assigned work change after the original record is created.

The methods

Photo identity document, checked against a live image by a person or by face matching.

Institutional login, which proves credential possession rather than identity.

Knowledge questions from student records.

Live challenge: showing a document, stating something, following an instruction.

And face matching against a previously enrolled image, which is biometric processing and has its own note.

What each actually proves

A document check proves somebody presented a document.

Login proves credentials were used.

Face matching proves similarity to a stored image, with an error rate that differs by group.

None of them is conclusive, and the appropriate strength depends on what the qualification is worth.

Where it matters most

Licensing and professional examinations, where an impersonated pass has consequences for the public.

Credit-bearing assessment that transfers between institutions.

And anywhere the qualification is a gateway rather than a record of learning.

For a formative quiz it is disproportionate, and applying it everywhere is a common overreach.

The biometric question

Face matching processes biometric data, which several jurisdictions treat as a special category requiring a stronger justification.

Storing a face template is a different commitment from storing a photograph.

Ask: is a template stored, where, for how long, and can a student require deletion after the assessment.

And whether a non-biometric alternative exists for students who object — which for identity checking usually does, since a human can compare a document to a face.

Accuracy by group

Face matching error rates differ measurably by skin tone, age and gender in published evaluations.

Which means a system used for identity will fail more often for some students, and those students will experience a challenge to their identity at the start of a high-stakes exam.

Ask the supplier for disaggregated figures. If they do not have them, that is the answer.

Practical failures

A student whose appearance has changed: hair, glasses, weight, gender presentation, medical treatment.

A student with no photographic identity document, which is more common than institutions assume.

Poor lighting in a bedroom at nine in the evening.

Each needs a human fallback that is quick and not humiliating.

What to check

What does your identity check actually prove?

Is biometric matching used, and on what legal basis?

Are error rates by group known?

And what happens to a student whose face does not match — how long does it take and who do they speak to?

The point

Face matching error rates differ measurably by skin tone, age and gender.

A system used for identity will challenge some students harder at the start of a high-stakes exam, and suppliers should be asked for disaggregated figures.

Worth stating

Ask whether a stored face template is stored, where, for how long, and whether a student can require deletion afterwards.

For identity checking a non-biometric route usually exists, since a human can compare a document to a face.

Also worth knowing

Impersonation is a different risk from assistance and the measures against it are different.

Conflating them produces systems that do both badly, and applying identity checks to formative quizzes is a common overreach.

And finally

A document check proves somebody presented a document, a login proves credentials were used, and face matching proves similarity with an error rate that differs by group.

None is conclusive, and the appropriate strength depends on what the qualification is worth.

Summary

Impersonation and assistance are different risks with different measures. Systems that try to do both usually do both badly.

In summary

Identity checking is a narrower problem than proctoring and deserves a proportionate answer.

Biometric matching on a weekly class test cannot be justified by its purpose. For wider institutional context, consult the European Commission.