AI interview proctoring software: what works in a live interview
Lockdown browsers and webcam monitoring were built for exams. Here's how interview proctoring software differs, and what actually works when the interview is a live video call.
The Candit Team, 2026-10-21
Interview proctoring software checks that a candidate's answers in a remote interview are their own. Most tools sold under that name were built for exams: they lock the browser, watch the webcam or record the screen. In a live video interview, that leaves the biggest gap open, because an AI copilot on a phone or a hidden overlay never touches what they watch.
The useful question isn't "which proctoring tool?" but "what does it watch?" A tool that watches the device can only see that device. A tool that reads the conversation sees the answers themselves, wherever they came from.
What is interview proctoring software?
Interview proctoring software is any tool that helps an interviewer trust that a remote candidate is answering unaided. There are two families: secure-environment tools that control and watch the candidate's computer, and conversation-based tools that read what's said in the call. Exams suit the first. Live interviews need the second.
The trouble is that search results blur them together. Many of the pages that rank for this term describe exam and assessment proctoring (tab switching, ID checks, face tracking) and apply the same toolkit to interviews, as if a timed test and a conversation were the same problem.
Why can't lockdown browsers stop interview copilots?
A lockdown browser controls one browser on one computer. It can't see a phone propped behind the laptop, and it can't see software built to stay out of the screen share. Universities that use one say so plainly: "LDB cannot detect or prevent students from accessing another device, or having someone beside them" (UBC Sauder Teaching Playbook).
Copilots are designed around exactly that blind spot. When CNBC profiled Interview Coder in March 2025, it reported that the tool's website called it "immune to screen detection features" on services like Zoom and Google Meet (CNBC via NBC New York).
That's why the live interviews we run to vet candidates don't watch the screen at all. Our interview tech joins the call, reads the conversation, and flags answers that look AI-assisted while the interview is still on, with the exact line behind each flag. A copilot can hide its window. The candidate still has to say the answer out loud.
How do the main approaches compare?
| Approach | What it watches | Install for the candidate | Copilot on a second device | Works during the call |
|---|---|---|---|---|
| Lockdown browser | One browser on one computer | Yes | Not seen | Yes |
| Webcam and screen monitoring | Camera feed and shared screen | Usually | Only if it's in frame | Often reviewed afterwards |
| AI notetaker | The conversation | No | Summarised, not judged | Notes only |
| Conversation-based detection | The answers themselves | No | Its answers are still spoken, so they're still read | Yes |
The friction matters too. In a 2025 University of Washington study of 16 students with disability accommodations who had sat proctored exams, one participant said: "It just feels like when you're being watched, you can't breathe" (Kwapisz et al., arXiv). It's a small sample of students rather than job candidates, but the point carries over: surveillance costs the honest candidates most.
Which interview proctoring tools use which approach?
Here's where the best-known tools sit, and what each says it uses on its own website.
| Tool | Approach | What it uses |
|---|---|---|
| Respondus LockDown Browser | Lockdown browser, built for exams | Locks one browser on the candidate's computer |
| Talview | Secure environment | A lockdown browser, a secondary camera, ID verification and live screen sharing |
| Capifiq | Device checks | Its own interview room, with a desktop client that checks the candidate's device for hidden AI tools, remote-control sessions, virtual machines and virtual cameras |
| Sherlock | Device and behaviour signals | Its own secure meeting links, with signals from device activity, the audio environment and candidate behaviour |
Device checks see more of one device, but only that device. Capifiq's own site says on-device detection can't see a phone, an earpiece or smart glasses, and handles those with interview questions instead. That's the gap conversation-based detection covers: whatever device the answer came from, the candidate still has to say it. For what "zero friction" should actually mean when vendors claim it, see zero-friction vs secure-environment proctoring.
Vendor details come from each company's own website, checked in October 2026.
What should interview proctoring software actually check?
Good interview proctoring software checks the answers, not the room. Look for four things: nothing for the candidate to install, detection that reads the conversation rather than the screen, evidence behind every flag, and flags that arrive while the interview is still on, early enough to ask a follow-up question.
It should also cope with how copilots try to dodge detection. Many now give the candidate bullet points to put in their own words. That changes the wording, not the shape: a pause while the points appear, then the points delivered in list order, with written phrasing slipping into speech. Conversation-based detection can recognise that format, because it reads the conversation and its timing, not the device.
And ask how the vendor knows it works: what it has been tested against, and how often it's retested as copilots update.
Why does this matter now?
Trust in video interviews is falling. In Checkr's survey of 3,000 US managers involved in hiring, fielded 11–16 September 2026, "Sixty percent say AI has made virtual interviews harder to trust, compared with 51% last year." Only 23% are "extremely confident their hiring process would catch a fraudulent candidate" (Checkr, October 2026).
The response many teams reach for is more surveillance, or a return to in-person rounds. Both add cost and friction, and neither tells you which answer to probe.
How we use it
We don't sell proctoring software. We're a recruitment agency, and conversation-based checks are part of how we vet every candidate before you meet them. Our interview tech joins the live interview on the candidate's normal video call, with nothing to install, and flags answers worth probing with the line that triggered them.
A flag is a signal, not a verdict. It tells the interviewer which answer is worth a follow-up; a person decides what happens next. Read more on real-time interview proctoring and how to tell if a candidate is using AI in an interview.
Frequently asked questions
Is interview proctoring the same as exam proctoring?
No. Exam proctoring controls and watches the test-taker's device for a fixed test. Interview proctoring has to work inside a live conversation, where the risk is a copilot feeding answers from somewhere the camera and screen share can't see.
Can Zoom, Teams or Google Meet detect AI copilots?
Not on their own. Meeting platforms show what's shared, and several copilots are built to stay out of the screen share or run on a second device. Detection has to read the answers themselves, as they're spoken.
Does conversation-based detection need access to the candidate's computer?
No. It reads the conversation and the call audio, with no access to the candidate's device, which is what keeps it zero-friction.
Does a flag mean the candidate cheated?
No. A flag marks an answer worth a follow-up question. It should come with the exact line behind it, and the interviewer decides.
Run interviews that hold up
If your team runs interviews over video, the fix doesn't need a new portal or a locked-down laptop. It needs to know which answers to probe while the candidate is still on the call.
Or hand the first round to us. We vet every candidate down to the interview before you meet them. Book a call.