Online proctoring privacy concerns, answered by the defaults

Online proctoring privacy concerns start with the defaults: on NUADU, an identity check and a lockdown browser; every other layer opt-in, with a DPIA template.

Most online proctoring privacy concerns come down to one question: which layers run, and who chose them. On NUADU the answer is deliberately narrow — an identity check at exam start and a lockdown browser by default, every other layer an explicit choice per assessment, no emotion or intent inferred about the candidate. This post takes the common concerns one at a time and answers each in a few lines, with a link to the post that covers it in depth. It is written for exam officers, and for the DPO they should bring.

Is online proctoring an invasion of privacy?

It can be, and what decides it is the set of layers switched on for a given exam. A camera and a recording during a weekly quiz are a different processing activity from a browser that blocks other tabs during a final. A programme that runs everything on every exam has stopped telling the two apart.

NUADU starts from the narrow end. For a new assessment, lockdown browser and identity verification at exam start are on; all other integrity layers are off by default and need an explicit choice per assessment. Recorded session review is opt-in. The additional environmental layers, reserved for the highest-stakes uses such as professional licensing and certification body exams, are off by default and require a DPIA.

The Conduct page is direct about why the narrow end matters. Over-reach has costs: test anxiety, accessibility problems, and regulator action against vendors that overreach. A defensible programme matches integrity controls to stakes — not the other way round. The invigilation and integrity layers on the Conduct page set out the full model.

Integrity is real. So is dignity. NUADU is designed around both.

Does proctoring software detect emotions, stress or attention?

Not on NUADU. The platform does not infer attention, stress, engagement, confidence or intent from a candidate's face, voice, gaze or keystrokes, and it performs no emotion recognition, intent inference, stress detection or attention scoring. Anomaly signals are objective events: a tab switch, an identity-mismatch flag, an unusual pattern.

The difference matters for privacy as much as for fairness. A tab switch is an event in the record that can be checked and disputed. An estimate of their stress or attention is a claim about their state of mind that they have no way to check.

In the EU, part of this refusal is also the law. The AI Act bans AI systems built to infer a person's emotions in education institutions, unless the system serves medical or safety purposes (Article 5(1)(f) of the AI Act). Integrity signals appear among NUADU's AI capabilities in assessment with the same pairing as every other module: a named invigilator reviews the anomalies, and a signal is never treated as a finding.

Who pays for a false flag?

The candidate does, and not every candidate equally. Among the costs of over-reach, the Conduct page lists false-positive flags that fall disproportionately on disabled, neurodivergent and non-white candidates.

A false flag is a privacy harm twice over: the candidate was watched, and then suspected. The design answer is to stop a flag from becoming a finding. On NUADU, integrity signals are surfaces for a named invigilator — not findings about a candidate — and the institution's own academic-integrity process decides what follows. The path of one flag, from the event to the decision and the candidate's route to contest it, is set out in what happens to proctoring flags.

Accessibility belongs in the same conversation: keyboard navigation and screen-reader labelling on interactive controls, and extra-time accommodations applied automatically, without re-verification overhead.

Is a lockdown browser an invasion of privacy?

Far less than a camera. A lockdown browser works on the exam session rather than on the room: on NUADU, browser controls lock the browser and detect tab switching, copy-paste and other applications, with severity configured per assessment. For most assessments, that layer is sufficient and proportionate.

It is not data-free. Each event is written to the session's audit trail, which is why a named invigilator reviews it and the institution, not the system, decides what it means. But a configuration can stop there: the illustrative classroom quiz on the Conduct page runs on browser controls alone — no cameras, no recording. How the higher rungs add layers, and why each one should be a choice, is laid out in the types of online proctoring.

How can institutions address online proctoring privacy concerns?

By deciding the controls per assessment, in writing, before the first sitting — and by keeping every judgement about a candidate with a named person. Five habits, each a decision the institution makes:

  • Start from the stakes: for many assessments, browser controls on their own are enough.
  • Verify, do not identify: a one-to-one match at exam start, or an alternative identity workflow — see biometric proctoring.
  • Treat recording as a processing activity with its own DPIA, never a default — see online proctoring under the GDPR.
  • Keep signals as signals: a named invigilator reviews, the institution decides, the system penalises nobody.
  • Ask for the paperwork before the first sitting: the DPIA template, the data flows, the retention policy and the list of sub-processors.

This is not the vendor's decision to make. The documentation pack is on request; for the DPA, data hosting and sub-processors, see GDPR compliance, data residency and the DPA. Integrity proportionate to stakes. Candidate due process intact.

→ See how Conduct works