Remote proctoring supervises candidates outside a conventional test centre, but its visibility is often narrower than buyers assume. Expert research demonstrates that low-cost off-device systems can use a separate camera, phone, auxiliary computer, or discreet feedback channel while leaving the monitored exam device apparently normal. AI-assisted monitoring can still support scale and triage, but it cannot by itself secure the candidate’s wider environment or repair a weak task. Buyers should compare remote delivery with redesign and controlled-centre options against the actual construct and stakes.

Remote proctoring and invigilation controls are the tools, workflows, and staffing models used to supervise assessments outside a traditional exam room. They may include identity checks, live human invigilation, recorded review, AI-assisted monitoring, secure browsers, room or environment checks, device detection, session logs, and incident review;. The key distinction is between monitoring the **assessment device** and assuring the **assessment environment**. A webcam and secure browser may observe the laptop, while a separate camera, phone, screen, wearable, helper, or concealed audio route remains outside that field of view.

Remote delivery can widen access and reduce dependence on venues, but it moves security work into the candidate’s home, device, network, and local environment. The quality of the result therefore depends on identity assurance, task design, monitoring, candidate support, review quality, and recovery when technology fails. The new expert evidence changes the procurement question. It is no longer enough to ask whether a system detects suspicious behaviour on screen. Assessment leaders should ask whether the model addresses line-of-sight cameras, auxiliary screens, hidden audio, wearable triggers, and other off-device routes without imposing disproportionate intrusion.

- **Live invigilation**: a human supervisor watches the session and may intervene in real time. - **Recorded review**: evidence is captured for later investigation. - **AI-assisted monitoring**: software identifies events or patterns for triage. - **Human adjudication**: a trained reviewer decides whether a signal warrants action. - **On-device assurance**: controls focused on the candidate’s main computer. - **Off-device assurance**: controls addressing other screens, cameras, phones, wearables, people, or devices. - **Candidate support burden**: the technical and administrative effort required to complete the assessment fairly.

Remote proctoring is a delivery layer, not a complete integrity solution. It can help with identity, session observation, or device access, but it cannot by itself repair weak assessment tasks, poor item security, or inadequate result governance;. Automated signals should normally support human review rather than become automatic findings of misconduct. Supplier positioning that combines AI monitoring with human oversight, privacy, compliance, and auditability is a useful description of the operating questions buyers should ask, but remains supplier-authored evidence. The expert research provides a credible reason to test the boundary of remote visibility directly. Cheap cameras, mobile phones, auxiliary computers, and discreet haptic or audio feedback can create an assistance route that leaves little evidence in the main session recording. This is a demonstrated capability and threat signal, not evidence of prevalence across all candidates.

The central contest is proportionality. AI-assisted monitoring may reduce routine review effort at scale, but it may also create opaque flags, accessibility problems, and new queues for human adjudication. Buyers should ask whether fewer visible interventions actually mean fewer false positives and better decisions. There is also a live choice between remote delivery and controlled centres. Centre delivery is not risk-free, but it may offer stronger environmental control for assessments requiring unaided recall. Remote delivery may remain defensible where the construct permits tools, where task design reduces the value of assistance, or where the stakes justify a carefully tested layered model.

- false positives from automated flags - false rejection or failure to authenticate legitimate candidates - off-device assistance that remains invisible to screen and webcam monitoring - accessibility barriers involving cameras, lighting, bandwidth, devices, or environments - privacy and data-retention concerns - operational overload from support, review, and appeals - candidate anxiety and construct distortion caused by intensive surveillance - overconfidence in supplier compliance or accuracy claims

1. **Define the assurance goal.** - Decide whether the control addresses impersonation, hidden assistance, off-device support, device access, environmental risk, or several of these. 2. **Map the complete candidate environment.** - Include auxiliary screens, phones, cameras, wearables, hidden audio, helpers, and objects with line of sight to the assessment screen. 3. **Red-team the model with ordinary equipment.** - Test inexpensive and plausible configurations, not only sophisticated cyberattacks. 4. **Choose the least burdensome effective model.** - Compare live human supervision, recorded review, AI-assisted triage, secure browsers, task redesign, and centre delivery. 5. **Specify human decision points.** - Document which events require review, who can make an adverse finding, and how candidates can challenge it. 6. **Measure what matters.** - Track false positives, false negatives where discoverable, authentication failure, appeals, support time, abandonment, and candidate impact separately from flag volume. 7. **Plan recovery and redress.** - Define what happens after an outage, wrongly handled flag, corrupted recording, or disputed identity decision.

| Model | Strength | Weakness | Best fit | |---|---|---|---| | **Human-led remote proctoring** | Strong contextual judgement and real-time intervention | Expensive and harder to scale; still has limited off-device visibility | High-stakes, lower-volume sessions with a tested environment policy | | **Recorded review** | Flexible staffing and a persistent evidence trail | Slower decisions and substantial storage/review needs | Sessions where post-event review is acceptable | | **AI-assisted monitoring with human review** | Scalable triage and consistent event capture | False positives, opacity, and off-device blind spots | Large cohorts with mature adjudication and red-team processes | | **Secure browser plus limited monitoring** | Lower intrusion for device-bound risks | Does not address off-device help or identity alone | Timed assessments with clear device rules | | **Test-centre delivery** | Stronger environmental control | Less flexible and more expensive to access | Critical decisions where residual remote risk is too high | | **Redesigned or tool-integrated assessment** | Protects authenticity by changing the evidence required | Requires revalidation and new marking rules | Assessments where generic answers are easy to outsource |

A certification provider receives a proposal for AI-assisted remote proctoring and assumes that the webcam and secure browser cover the main risks. A red-team exercise tests a separate visual feed and discreet feedback outside the exam laptop. The provider then decides that its highest-stakes knowledge test should move to a controlled centre, while lower-stakes or applied assessments are redesigned around explanation and judgement.

- Expert research presentation demonstrating low-cost off-device cheating-system capabilities. - Overview of live, recorded, and AI-assisted models. - Supplier description of AI-assisted monitoring and human oversight. - Presentation note on candidate journey mapping. - Comparison of detection-led and design-led responses.

The market includes live proctoring providers, AI-assisted monitoring platforms, secure-browser suppliers, identity-verification services, device-detection tools, and assessment platforms that bundle several controls. Suppliers commonly frame their offers around scale, compliance, privacy, auditability, and human oversight. The new research means buyers should add off-device coverage to their evaluation. Ask whether the supplier has tested auxiliary cameras, separate screens, phones, wearables, concealed audio, and ordinary household objects with line of sight to the test screen. A product may be effective at analysing the monitored device while leaving the main threat untouched.

### Is remote proctoring enough on its own to make an exam secure? No. It is one delivery-layer control and may not address off-device assistance, weak task design, item leakage, or result governance. ### Can a secure browser detect a separate camera or phone? Usually not simply because the browser is locked down. Those devices may operate outside the exam computer and need a different threat response. ### Is AI proctoring always better than human proctoring? No. AI can support scale and triage, while humans remain important for ambiguity, context, sanctions, and appeals. ### What should I ask a proctoring supplier first? Ask what exact risk the product reduces, whether it covers off-device routes, what independent evidence supports the claim, how false positives are handled, and who makes the final decision. ### Should I move an assessment into a test centre if remote cheating is possible? Not automatically. Compare the stakes, construct, candidate access, residual risk, and the feasibility of redesign. Centre delivery is more compelling when unaided knowledge is essential and remote controls cannot provide credible environmental assurance.

Tim Burnett (Admin)

`Test Community Network. "Remote proctoring and invigilation controls." TCN AI & Assessment Wiki. Last reviewed 2026-09-28. https://testcommunity.network/knowledge/test-security/wiki/remote-proctoring-and-invigilation-controls`

- Expert research presentation on off-device cheating systems. - Source page on remote supervision models. - Supplier claim on AI-assisted monitoring and human oversight. - Presentation note on candidate journey mapping. - Comparison of detection and redesign.