By Oliver Bennett, MSc in Work and Organizational Psychology
In short
Remote proctoring can observe a browser's state, a camera frame and a screen's contents, and nothing outside those three. A second person off-camera, a phone below the desk, prior knowledge of the questions and work done before the session all fall outside what it can detect. This makes it effective against casual, in-session improvisation and weak against anything prepared, which is the opposite of the threat model most buyers have in mind.
Proctoring is almost always evaluated by what it catches. Vendors demonstrate detections; buyers ask about accuracy; the conversation stays inside the set of things the system can see.
The more useful inventory runs the other way. What can a proctored session structurally not reach? That list is short, stable, and it decides which assessments the technology actually protects.
Three surfaces, and a boundary
Everything a remote proctoring system knows comes from three places.
The browser. Focus changes, copy and paste, print attempts, an additional display being connected, the window leaving fullscreen. These are real, accurately recorded, and confined to the tab.
The camera frame. Whatever a forward-facing lens contains, at the resolution and interval configured.
The screen. What is displayed on the monitor being captured.
Those three define a boundary, and the boundary is the whole subject. A system cannot observe what is not in a browser event, a camera frame or a screen capture — not because of implementation quality, but because there is no sensor pointed at it.
What falls outside
A second person. Someone seated beyond the camera's angle, reading the screen and speaking. Nothing in the three surfaces contains them. A room scan at the start establishes the room at the start, which is a weaker claim than it is usually read as.
A second device. A phone below the desk, a tablet beside the laptop, a second machine entirely. The screen capture covers the captured screen.
Preparation. Anything done before the session begins — a prepared answer bank, a memorised response, material written out in advance — arrives as ordinary typing.
Prior knowledge of the items. The largest gap, and the one least discussed. A candidate who has seen the questions answers them correctly under full observation, and the session looks exemplary throughout.
| Threat | Observable in session |
|---|---|
| Improvised lookup in another tab | yes — focus change, paste |
| Copying the question text out | yes — copy event |
| A second person off-camera | no |
| A phone below the desk | no |
| A prepared answer typed in | no |
| Having seen the items beforehand | no |
Read down that column. Proctoring is strong against unplanned, in-session improvisation and weak against anything prepared — which is the inverse of the threat model most buyers describe when asked what they are worried about.
The asymmetry that follows
Someone improvising has made no arrangements and leaves traces in the browser. Someone who planned has arranged exactly the conditions the three surfaces do not cover.
So proctoring catches the opportunistic candidate and not the determined one. That is still worth something — the opportunistic group is larger — but it inverts the usual justification, which is framed around the serious cheat and the high-stakes exam.
And the cost of the control does not scale with the threat. Every honest candidate pays the same attention tax, friction and tone of suspicion, regardless of which group the monitoring is aimed at.
Item exposure is the unaddressed problem
If a question has circulated, no amount of session monitoring restores what the score was supposed to mean. The candidate is answering a question they have already seen, correctly, while being watched closely.
This is a content problem with content solutions: larger item pools with randomised draws so no two candidates see the same set; items rewritten on a schedule rather than when a leak is noticed; and tasks built on specific cases rather than reusable general ones, since a case-specific task is harder to circulate usefully.
Those measures do more for the validity of a score than any configuration of monitoring, and they are invisible to candidates — which means they cost nothing in experience.
What to ask before enabling anything
Two questions, in this order.
What would a successful cheat actually gain? For a thirty-minute screening task that feeds a work sample and two interviews, the answer is usually: a stage further into a process that will catch them. That containment is itself a control, and it means the early stage can stay light.
What is the honest cost? Not only the licence. Candidates who withdraw when they see a webcam requirement on a screening task, the anxiety that degrades performance on precisely the dimensions a written work sample measures, and the reputational cost of every false accusation — each of which becomes a story that travels.
Where the stakes genuinely justify it — a regulated certification, a credential with standing — proctoring earns its place, and the limits above are reasons to combine it with item management rather than reasons to skip it.
The strongest signals need no camera
Worth ending here, because it reframes the spend.
The candidate's work is usually the best integrity evidence in the file. A response that does not engage with the specific details of the brief gives itself away. A response that addresses every particular of the case is hard to fake, whatever the session log says. Reading the work costs a reviewer minutes and requires no surveillance at all.
That is not an argument that proctoring is worthless. It is an argument about order: design the task so it is hard to outsource, manage the items so they are not already known, read the work, and add monitoring where what remains at stake still justifies it. Most processes do these in reverse, and buy the camera first.
- proctoring
- remote assessment
- assessment integrity
- surveillance
- test security
Key takeaways
- Three observation surfaces — browser, camera frame, screen — define the boundary. Everything outside the frame is outside the system.
- Proctoring is strongest against unplanned, in-session lookups and weakest against preparation, which is the more consequential threat.
- Nothing in a proctored session addresses item exposure: questions that have circulated are answered correctly by candidates who are being watched closely.
- The cost is paid by every honest candidate, which is why the question is never 'does it catch anything' but 'does it catch enough to justify that'.
Sources & further reading
- Nigam, Pasricha, Singh & Churi (2021), A Systematic Review on AI-based Proctoring Systems — Education and Information Technologies
- Almossa & Alzahrani (2022), Lessons on Maintaining Assessment Integrity — International Journal for Educational Integrity
- AERA, APA & NCME — Standards for Educational and Psychological Testing