Assessment with Proctoring: Organising Secure Remote Exams

A remote exam without active supervision is not a secure exam. It is an exam on the honour system.

Academic fraud is on the rise. The UK’s Quality Assurance Agency has documented a significant increase in academic misconduct cases since the shift to remote assessment. Generative AI tools have added a new layer: a candidate can obtain sophisticated answers in seconds, without leaving their screen.

Proctoring is the structured response to this problem. Not a perfect solution – none exists. But a framework that deters, detects, and documents, where the absence of supervision does none of these things.

This article explains what proctoring really is, how to choose the right level according to stakes, what it implies from a data protection standpoint, and how to deploy it without creating more problems than it solves.

---

1. What Proctoring Really Is

Definition and Scope

Proctoring refers to the set of technical and human measures put in place to supervise the conduct of a remote exam and guarantee the identity of the candidate, the integrity of their environment, and the authenticity of their responses.

It is not just a webcam turned on. A complete proctoring framework covers multiple layers: identity verification on entry, IT environment security, behavioural monitoring during the assessment, and production of archivable evidence afterwards.

Each layer plays a distinct role. Removing one creates a vulnerability. A candidate verified on entry but whose screen is not monitored can consult their notes freely. AI proctoring without identity verification does not guarantee the right candidate is actually sitting at the computer.

The 3 Levels of Proctoring

Proctoring is not a single technology. It is a spectrum of solutions, with very different security levels and operational costs.

Browser lockdown is the basic level. It blocks access to other tabs, applications, and the internet during the assessment. Simple to deploy, low cost, but limited: a candidate with a second device (phone, tablet) can work around it undetected.

AI proctoring adds a behavioural surveillance layer via webcam and screen recording. Algorithms analyse gaze movement, tab switching, the presence of a third person or second screen. They generate an incident report reviewable after the exam. AI flags anomalies – it does not prove fraud. The decision remains human.

Live proctoring deploys a human invigilator in real time, via video call. This is the highest level of security – and the most costly. The typical ratio is 1 invigilator per 8 to 15 simultaneous candidates. It is justified for very high-stakes exams: competitive admissions, professional certifications susceptible to challenge, competitive selections.

What Proctoring Measures – and What It Does Not Guarantee

Proctoring reduces fraud risk. It does not eliminate it.

A determined candidate can memorise their notes, use a phone off-camera, or cheat with an accomplice out of shot. Proctoring acts on two levers: deterrence (visible surveillance discourages most attempts) and documentation (in the event of proven fraud, recordings constitute admissible evidence).

What it does not do: guarantee the candidate is alone, that they have not prepared handwritten notes off-camera, or that they are not using an AI tool on an unsupervised device.

---

2. AI Proctoring vs Live Proctoring: How to Choose

The 3 proctoring levels: how they work and recommended use cases

Level	How it works	Fraud detection	Operational cost	Recommended use
Browser lockdown	Blocks access to other tabs and applications	Low (bypassable with a second device)	Very low	Continuous assessment, low stakes
AI proctoring	Algorithms analyse behaviour via webcam and screen	Medium (flags anomalies, does not prove)	Low to medium	End-of-module exams, volume recruitment
Live proctoring	Human invigilator in real time via video call	High	High (1 invigilator / 8-15 candidates)	Competitive exams, high-stakes certifications

AI Proctoring: Advantages and Limits

AI proctoring is scalable. It can simultaneously supervise thousands of candidates without proportional cost increase. For an organisation managing high-volume recruitment campaigns or multi-site exam sessions, it is often the only economically viable model.

Its known limits: algorithms generate false positives (a glance to the side interpreted as cheating) and false negatives (well-concealed fraud undetected). Alerts must be reviewed by a human before any sanction. Using the AI report as direct evidence, without human review, exposes the organisation to challenges.

Live Proctoring: When It Becomes Indispensable

Live proctoring is justified when the exam stakes are high enough that undetected fraud would have significant consequences: certification invalidation, competitive exam challenge, institutional legal liability.

A professional certification board, an admissions competition at a prestigious institution, a selection exam for a regulated position: these are cases where the cost of live proctoring is justified by the cost of undetected fraud.

Hybrid Approach for High Stakes

The most effective combination: continuous AI proctoring during the assessment, with a human supervisor intervening on the most critical alerts in real time. AI filters the noise – the human supervisor decides on ambiguous cases.

This hybrid approach reduces the operational cost of full live while maintaining a high security level. This is the positioning of TestWe Remote Supervision: 128,552 supervised assessments facilitated in 2024, with a framework combining algorithmic detection and human intervention on incidents.

---

3. Data Protection and Ethical Considerations

Personal Data, Consent, and Legal Basis

Proctoring collects sensitive personal data: video recordings, screenshots, behavioural data. Depending on the level of surveillance, it may involve biometric data (facial recognition for identity verification).

GDPR requires an explicit legal basis for this processing. Generally, this is the contract (the exam is part of a training or certification programme the candidate has enrolled in) or consent (collected before the session). Consent must be freely given – a candidate who « consents » under pain of exam exclusion is not giving free consent under GDPR.

The organisation must inform candidates before the assessment: what data is collected, for how long it is retained, who has access, and what their rights are.

Data Hosting

Video recordings from exam sessions are personal data. Their hosting outside the European Union without an adequate transfer mechanism (standard contractual clauses, adequacy decision) is unlawful.

This is a systematic verification point when choosing a platform: where is video data hosted? For how long is it retained? Who can access it and under what conditions?

Regulatory Guidance on Remote Exam Supervision

Data protection authorities across Europe – including the ICO in the UK and GDPR supervisory authorities in EU member states – have issued guidance on remote exam surveillance. They consistently emphasise the principle of proportionality: the level of surveillance must be appropriate to the stakes of the exam.

Deploying biometric proctoring for a low-stakes formative assessment is not proportionate. Regulators recommend considering alternative measures first: question randomisation, broad question banks, open-ended questions that are difficult to copy. Proctoring should be a last resort, not the first.

---

4. Deploying Proctoring: What to Anticipate

Candidate Side: Information and Technical Preparation

A successful proctored exam begins with proper candidate preparation. The most frequent technical incidents – unrecognised webcam, incompatible browser, too-slow connection – can be anticipated with a preliminary test session.

Candidate information before the assessment must cover: required equipment (webcam, microphone, stable connection), environment rules (quiet room, clear desk, alone in the room), data collected and associated rights, and the procedure in the event of a technical incident.

A test page accessible 48 to 72 hours before the assessment validates technical compatibility and significantly reduces incidents on exam day.

Organisation Side: Configuration and Incident Review

Configuring AI proctoring requires calibrating alert thresholds. Thresholds set too sensitively generate hundreds of false positives to review manually – eliminating the time saving. Thresholds set too loosely allow suspicious behaviour through.

The incident review process must be defined before the exam: who is responsible? Within what timeframe are alerts addressed? What decision can be taken on the basis of an AI alert alone, and what decision requires further investigation?

Without a defined review process, incident reports accumulate without processing – and lose their evidential value.

Handling Challenges and Probative Value of Logs

The real value of proctoring is not only in detection. It is in documentation. A timestamped log tracing every candidate action – connection, screen activity, triggered alerts, invigilator interventions – is usable evidence in the event of a challenge.

Without this structured documentation, the organisation cannot defend its decisions. With it, it can justify a sanction or, conversely, demonstrate that an accused candidate did not in fact cheat.

---

5. Proctoring Across Contexts

Professional Certification: Auditability and Evidence

In professional certification, proctoring meets an auditability requirement. A certification body must be able to prove, in the event of an audit or jury challenge, that the assessment was conducted under conditions guaranteeing the authenticity of results.

Session logs, incident reports, and video recordings constitute the evidence file. For more on complete exam organisation, see our guide on organising professional certification assessments.

High-Volume Recruitment: Equity and the EU AI Act

In recruitment, proctoring ensures all candidates sit the assessment under the same conditions – without any candidate benefiting from external assistance that others did not have.

When AI tools intervene in the selection process (automated scoring, algorithmic proctoring), the organisation falls within the scope of the EU AI Act. Traceability and human oversight obligations apply. For more, see our article on test-based recruitment and selection objectivity.

Higher Education: Competitive Exams and End-of-Year Assessments

In higher education, the proctoring level must be appropriate to the stakes of the assessment. A low-weight continuous assessment does not require the same guarantees as an admissions competition or a final year exam determining degree award.

The proportionality rule applies: the higher the stakes, the more surveillance is justified. See our article on online exams in higher education for more on mixed frameworks.

---

Conclusion

Proctoring is not a universal answer to exam fraud. It is a graduated framework to calibrate according to exam stakes, candidate volume, and regulatory constraints.

Five key takeaways:

• Three levels, three use cases: lockdown for low stakes, AI for volume, live for high stakes
• AI proctoring flags, it does not prove: the sanction decision remains human and must be documented
• GDPR/data protection applies in full: legal basis, EU hosting, retention periods, candidate rights
• The value of proctoring is in documentation: logs and reports are the evidence in the event of a challenge
• Surveillance must be proportionate: data protection regulators recommend adapting the framework to the actual stakes of the exam

To deploy a remote supervision framework suited to your assessments – certification, recruitment, or higher education – TestWe combines AI and human proctoring, with native traceability and European data hosting.

---

FAQ