Biometric proctoring: verification, not identification

Biometric proctoring on NUADU: one face, one enrolled photo, one reviewable match at exam start. What the DPO's DPIA needs, and four options, including off.

Biometric proctoring is the use of a physical characteristic, such as the face, to check who is sitting an online exam — and the phrase covers two very different processing activities. On NUADU it means one of them only: at exam start, the candidate's face is matched once against the photo they enrolled, with a reviewable match decision. It is not continuous identification and not facial-recognition matching against a database. That distinction changes the DPO's proportionality argument. It does not remove the need for a DPIA, and there is an off-switch.

What is the difference between identification and verification in biometrics?

Identification asks who someone is and answers by searching one face against many stored faces: one-to-many. Verification asks whether someone is the person they claim to be and answers by comparing one face with one stored record: one-to-one.

The AI Act draws the same line in its definitions. Biometric identification compares a person's biometric data with the data of individuals stored in a database; biometric verification is the automated, one-to-one check of an identity against biometric data the person provided earlier (Article 3, points (35) and (36), of the AI Act). The Act's high-risk category for remote biometric identification then expressly excludes systems whose sole purpose is to confirm that a person is who they claim to be (Annex III, point 1(a)) — an exclusion from that category, not from the rest of the Act.

In an exam, verification has a narrow purpose: confirming that the candidate at the keyboard is the candidate on the register. Identification would mean finding out who a person is by searching everyone stored, which is not a question an exam needs answered.

Identity verification — not identification. Not facial-recognition matching against a database.

Does biometric proctoring use facial recognition?

On NUADU, only as a single one-to-one match at exam start — never as facial-recognition matching against a database, and never as continuous identification through the session. One comparison, once, with a decision a person can review.

For a new assessment, this match and a lockdown browser are the only defaults; everything else waits for an explicit choice. When the face and the enrolled photo do not match, the result is an identity-mismatch flag, one of the integrity signals the Conduct page lists, and it goes to a named invigilator rather than to an automatic penalty. Nothing else is read from the face: no attention, no stress, no intent.

Recorded session review is a separate layer with its own question to answer: off by default, opt-in per assessment, and the subject of a different DPIA conversation. The layered approach to invigilation on the Conduct page shows where each layer sits, and the wider online proctoring privacy concerns — cameras, recording, inference — are answered in one place.

Is face verification special category data under the GDPR?

A DPIA should start from the assumption that it is. The GDPR defines biometric data as personal data from specific technical processing of a person's physical characteristics, facial images among them, that 'allow or confirm' that person's unique identification (Article 4(14) GDPR) — and confirming identity is what a one-to-one match does. Biometric data processed to uniquely identify a person falls under the special-category rules of Article 9, which prohibit such processing unless one of the conditions listed there applies.

So the verification–identification distinction does its work inside the DPIA, not instead of it. Article 35 asks for an impact assessment before any processing likely to put people's rights and freedoms at high risk, and the Higher education page describes the moment it gets skipped: a DPO asked to sign off on biometric proctoring without a DPIA. Three points are specific to the biometric step:

  • Special category: which Article 9 condition the institution relies on, settled with its DPO before the first sitting.
  • Necessity: whether this assessment needs a biometric step at all, given that browser controls alone are sufficient for many assessments.
  • Proportionality: one-to-one, once, at exam start, with a named invigilator for any mismatch — weighed against a human ID check.

The rest of the assessment follows the step-by-step walk in online proctoring under the GDPR. The DPIA remains the institution's document; NUADU's part is the material for it: template, data-flow diagrams, retention policy, sub-processor list.

Can a candidate sit a proctored exam without biometric verification?

Yes. Where biometric verification is not preferred, identity verification at exam start can be replaced with an alternative identity workflow — a proctor in the room, or a government-ID check by an invigilator — and browser controls alone are sufficient for many assessments. Identity can run four ways:

  • Default: a one-to-one match against the enrolled photo at exam start, reviewable, not continuous.
  • Opt-in layer: government-ID matching, for assessments whose stakes justify it.
  • Alternative workflow: a proctor in the room, or a government-ID check by an invigilator.
  • Off: browser controls alone, with no identity step.

Each is a per-assessment choice, and the DPO should expect to see it written down that way. The decision belongs to the institution, not the vendor, and no biometric step at all is a legitimate rung, not a workaround. Accommodations deserve the same reading: integrity settings should be checked against them, and an extra time accommodation is applied automatically, without re-verification overhead.

No accuracy figure, no retention period

This post quotes no match-accuracy figure, and a DPO should treat any figure offered without its conditions with care. NUADU does not claim the step is infallible. It claims that the match decision is reviewable by a named person and that a mismatch is a signal, not a verdict.

Sub-processors and retention are answered in documents rather than in a blog post. The sub-processor list is provided with the DPA and updated on change, retention is configurable per data category, and the retention policy sits in the documentation pack — all on request, with the posture summarised under data residency and the DPA on the Trust page.

The mechanism is deliberately plain: one face, one enrolled photo, one comparison at exam start, one named reviewer for a mismatch. That is a sentence a DPO can put in a DPIA, a works council can read and a candidate can understand.

→ Walk through the layered model with our team — and bring your DPO