Proctoring is the wrong answer to a real problem

When interviews moved remote and AI assistance became ubiquitous, a familiar reflex kicked in: watch the candidate more closely. Turn on the webcam. Track the eyes. Time the keystrokes. Flag the tab switch. Record the room. The logic feels airtight: if you cannot trust the environment, monitor it. But proctoring imports a set of problems that are worse than the one it claims to solve.

Behaviour is not evidence of competence or deception

Proctoring rests on an assumption that does not survive contact with real people: that behavioural signals reveal intent. They do not. A candidate looks away from the screen because they think, not because they cheat. They pause because the question is hard, not because they are reading an answer off a second monitor. They type in bursts because that is how they write, not because they are pasting. Every one of these signals has an innocent explanation that is more common than the guilty one.

The result is a flood of false positives aimed at exactly the people you least want to lose. Anxious candidates fail proctoring. Neurodivergent candidates fail proctoring. Candidates with cheap webcams, poor lighting, shared housing, or caregiving interruptions fail proctoring. None of these correlate with competence. All of them correlate with being a normal human in a non-ideal room.

The legal and trust costs are not hypothetical

Behavioural and biometric monitoring sits in the path of a growing body of regulation. Biometric data carries specific consent and retention obligations in many jurisdictions. Automated decisions about people trigger additional rights and disclosures. A proctoring flag that contributes to a rejection is exactly the kind of automated, opaque, behaviour-based judgment that regulators have started to scrutinise. The compliance surface you take on is large, and it grows.

The trust cost arrives sooner. Senior candidates, the ones you most want, have options. Asking a staff engineer with three offers to install monitoring software and sit under a webcam tells them exactly how your company treats people. Many will simply walk. You will have optimised your funnel against fraud and, in the process, against your strongest applicants.

The real problem is real

None of this means the underlying worry is imaginary. It is genuinely harder to tell, from a polished remote conversation, whether a candidate can do the work. That problem deserves a serious answer. Proctoring is just not it, because proctoring answers a question about behaviour when the question you care about is about capability.

The better answer is to make the content of the interview do the work. If your questions are anchored to the specific claims on a candidate's CV, and they probe the reasoning behind those claims, then it does not matter what the candidate does with their hands or eyes. Someone who led the migration can walk you through the incident that followed it. Someone who did not will run out of detail, no matter how calm their webcam feed looks.

Score the answer, not the nervous system.

Structure the evidence, keep the human

There is a version of interview rigour that does not require surveillance. The interviewer asks claim-anchored probes, then marks which evidence signals actually appeared in the answer: did the candidate name the failure mode, explain the trade-off, show ownership of the decision. That produces a structured, comparable record across candidates for the same role, which is the legitimate goal proctoring was reaching for. It just gets there by reading the content of what was said, not by watching the person say it.

Crucially, the interviewer still decides. The tool never scores, ranks, or rejects a person. It helps a human ask sharper questions and record what they observed. That is the line: structure the evidence, never the behaviour. Hold it, and you get a screen that is both fairer to candidates and sharper for you. Cross it, and you get a surveillance system that punishes the wrong people and exposes you to risk you did not need to take.