AI-enabled assessment security is about keeping assessment trustworthy when AI makes it easier to generate answers, conceal assistance, or disguise identity. It also covers the use of AI and biometrics for identity verification and remote proctoring. The core issue is not just misconduct detection; it is whether task design, invigilation, device policy, and identity controls still produce valid evidence when AI support is cheap, discreet, and embedded in ordinary tools. The source set points towards tighter authenticity design and stronger controls, but also shows that more surveillance does not automatically mean better assessment.

AI-enabled assessment security refers to the design, policy, monitoring, and identity controls used to protect assessment integrity in contexts where AI can support cheating or identity disguise. It includes traditional security concerns such as device restrictions, invigilation, and authentication, alongside newer AI-based tools such as remote proctoring, typing biometrics, facial recognition, and other identity checks. The deeper assessment question is whether the work still evidences the learner’s own capability under realistic conditions.

Security failures can undermine validity, fairness, public trust, and the credibility of results. Where AI support is difficult to detect or easy to use, assessment teams need to look beyond misconduct rules and examine authenticity design, exam-room controls, identity verification, privacy, accessibility, and whether the policy can actually be enforced. AI-based proctoring and identity systems also raise a practical question: do they meaningfully improve assurance, or do they mainly add another layer of monitoring and vendor dependence?

- **Authenticity**: whether the task still evidences the learner’s own capability. - **Control**: whether device rules, invigilation, monitoring, and authentication are proportionate and enforceable. - **Design**: whether the assessment should be redesigned rather than only policed. - **Governance**: whether policy, privacy, accessibility, and appeals routes are clear enough for high-stakes use.

The strongest evidence here points to a simple convergence: AI is increasing pressure on assessment security from two directions at once. It can support candidates directly during coursework or live exams, including through phones or wearable devices, and it is also being marketed into identity verification and remote proctoring as part of the control response. Regulator-facing reporting suggests this is already an operational issue rather than a hypothetical one, with Ofqual asking exam boards for stronger action on AI misuse and mobile phones in exam halls. That points towards a real shift in how security is being discussed: not as a narrow misconduct problem, but as a design and governance problem too. The evidence also supports a warning that ordinary components can be combined into covert support channels. The smart-glasses demonstration is best read as a threat illustration rather than evidence of prevalence, but it helps explain why assessment teams cannot rely on old assumptions about what is visible in the room or on the screen. Vendor pages consistently cluster around typing biometrics, facial recognition, document automation, room scans, OTP, randomised questions, and continuous monitoring. That is a useful market signal, but it remains vendor framing unless independently validated in assessment settings. It does not settle whether the controls are accurate, proportionate, accessible, or robust across candidate groups.

The main open question is which controls actually improve validity and security without creating new fairness, privacy, or accessibility problems. The source set is stronger on identifying risk than on proving which specific control stack works best across assessment types. Another unresolved issue is whether some assessments should be redesigned rather than defended with more surveillance. The evidence suggests a recurring tension: stronger monitoring and authentication may improve control, but they do not automatically produce better assessment if the underlying task still invites unauthorised support or if the controls themselves distort access and performance.

- AI-assisted misconduct can weaken trust in results and the fairness of certification. - Over-reliance on detection or surveillance can create privacy, accessibility, and proportionality problems. - Identity systems and remote proctoring can fail legitimate candidates if error handling is weak. - Procurement can drift towards tools that monitor more but explain less. - Assessment redesign may be postponed because surveillance feels like an easier fix.

1. Define the threat you are actually managing: authorship, impersonation, unauthorised support, or all three. 2. Check whether the task still needs unaided evidence, or whether AI should be permitted in some form. 3. Match controls to the stakes and the evidence you need. 4. Test accessibility, false positive risk, and candidate experience before scaling surveillance. 5. Use monitoring as one part of a wider authenticity strategy, not as a substitute for it. 6. Ask suppliers for evidence in live conditions, not just feature descriptions.

| Option | What it means | Main strength | Main concern | |---|---|---|---| | Minimise surveillance | Rely more on assessment design and authenticity controls | Lower privacy and accessibility burden | May leave some risk less directly managed | | Use targeted controls | Apply monitoring only where the risk justifies it | More proportionate than blanket monitoring | Needs clear rules and review routes | | Integrate broad surveillance | Embed monitoring across platforms and workflows | Consistent control and easier operational rollout | Risk of overreach and candidate distrust |

A programme team notices increasing concern about take-home work completed with AI support. Rather than adding blanket proctoring, it first reviews which tasks actually need unaided performance evidence, which can permit limited tool use, and where post-hoc monitoring is proportionate. In this scenario, surveillance becomes one part of a wider authenticity strategy rather than the default answer.

- FE Week / Schools Week article on Ofqual and AI misuse. - Smart-glasses demonstration article on AI-assisted exam answering. - TypingDNA identity assurance vendor page. - Verif-y identity verification vendor page. - Wheebox remote proctoring vendor page. - Witwiser proctoring/assessment vendor page. - Xobin remote proctoring vendor page.

The vendor landscape is centred on remote proctoring, identity verification, webcam monitoring, analytics, and device integration. These pages are useful for understanding how suppliers frame the problem, but they should be read as market signals unless independently validated. The market emphasis appears to be on more surveillance-enabled assurance rather than on reducing the need for surveillance through assessment redesign.

### What is digital surveillance in higher-education assessment? It is the use of monitoring, logging, remote proctoring, device control, and platform telemetry to support assessment integrity. ### Does digital surveillance make assessment more secure? It can improve oversight, but it does not solve authenticity or misconduct risks on its own. The deeper question is whether the surveillance is proportionate and whether assessment design should change as well. ### What are the main concerns with remote proctoring? The main concerns are false positives, privacy, accessibility, candidate anxiety, and whether humans have clear oversight and review routes. ### What evidence is still needed? Independent work on false-positive rates, accessibility impact, candidate experience, and whether surveillance actually improves authenticity in high-stakes assessment.

Tim Burnett (Admin)

Test Community Network. "Digital surveillance in HE assessment." TCN AI & Assessment Wiki. Last reviewed 2026-04-22. https://www.testcommunity.network/wiki/digital-surveillance-in-he-assessment.html

- Test Community Network conversation with Sarah Grayston on digital surveillance in higher-education assessment. - DigiProctor vendor source. - Talview vendor source. - MeritTrac vendor source. - Mobile Testing Solutions vendor source. - Owlya vendor source.