Second screen cheating in remote exams: the phone below the webcam
What the camera cannot see is most of the room
A webcam frames a face and a sliver of background. Below and beside that frame is an entire unobserved workspace. Proctoring software controls the exam device: lockdown browsers, process monitoring, screen recording. It has no authority over the phone in the candidate's lap, and no reliable way to know it exists. Room scans at check in see a desk before the exam starts. They say nothing about the device that arrives at minute twenty.
Vendors sell detection layers on top of this geometry: gaze estimation, head pose, audio analysis for second voices or keyboard patterns, phone shaped object detection in frame. Each catches a slice. Each also produces false flags, and the flags land disproportionately on candidates with disabilities, shared housing, or older hardware, which is a fairness problem with legal edges, not just a statistics problem.
Design answers that survive contact with a phone
If answers can be found on a phone in ninety seconds, the item is the vulnerability. Programmes reducing second screen value rewrite items toward application: novel scenarios, data the candidate must interpret rather than recall, multi step tasks where a searched fact is one ingredient among several. Timed sections sized so that searching costs more than knowing, and randomised item pools so a helper's leaked set helps one candidate, not a cohort, cut the return on the device under the desk.
Oral follow ups and short viva sampling on a percentage of candidates change the economics completely. A candidate who must explain their answers live, briefly, after the exam cannot outsource the understanding, only the typing, and the threat of the viva disciplines the whole cohort at the cost of examining a sample.
Handling flags like an institution, not a trap
A gaze flag is a lead, not a verdict. Build the review so a human sees the whole session context, the flag density, and the candidate's explanation before any accusation is written, and keep the accusation language to what the evidence shows: behaviour inconsistent with the rules, reviewed on these recordings, on these dates. Candidates appeal, regulators ask, and courts in some jurisdictions have already ruled on how far room surveillance may go.
The programmes that hold up are the ones that can show proportionality: the least intrusive controls that protect the credential, applied evenly, with a human accountable for the final call. The phone under the desk is real. So is the candidate. A defensible system has to see both.
Related reading: How do virtual machine tricks let test takers cheat inside locked-down exam browsers?, How do collusion rings coordinate cheating during live online exams?.
Frequently asked questions
Can proctoring software detect a phone outside the camera view?
Not directly. It infers from gaze direction, head movement, audio, and behaviour patterns, and those inferences are probabilistic. Some systems use a second camera, often the candidate's own phone placed to view the workspace, which genuinely closes the blind spot but raises its own privacy and access questions that programmes must justify and support.
Is looking down during a remote exam proof of cheating?
No. Candidates look down to think, to use permitted scratch materials, to rest their eyes. Gaze data only means something in pattern and context, reviewed against the session as a whole. Treating a single downward glance as evidence is how institutions lose appeals and credibility at the same time.
Why not just ban remote exams if second screens exist?
Because access is the counterweight. Remote assessment serves candidates who cannot travel to centres: disabled candidates, remote workers, entire regions without test centres. The professional task is to design assessment whose value survives a phone in the room, not to withdraw access from everyone because a minority cheat.
What is the strongest single control against second screen use?
Assessment design, specifically items that require applying knowledge to unseen material, backed by randomised pools. Surveillance raises the risk of being caught. Design lowers the value of the cheat itself, and it does so without collecting one extra frame of anyone's home.