Remote interviews have a cheating problem. Someone else answers, a second screen feeds the answers, or one person applies under five names. The tempting fix is software that watches faces and voices for "suspicious behavior." We think that fix does more harm than the cheating.
AI interview cheating detection that stays fair
There are two kinds of signal. Facts about the session: the candidate left the page, text was pasted, a phone appeared on camera. And guesses about the person: their eyes moved too much, their voice sounded nervous, their answer "reads like AI." Facts can be checked. Guesses punish people for being nervous, neurodivergent, or speaking a second language. Motium uses facts only.
What Motium checks
- Who is interviewing. Before the interview the candidate photographs a government ID and takes a selfie, checked by Stripe. If they have no ID, the hiring team confirms who they are another way.
- Leaving the page. When the candidate leaves the interview page, and for how long.
- Pasting. When text is pasted into the interview, and how much.
- Someone or something else on camera. A phone, another screen or another person in view.
- Screen sharing, when the position needs it. Some positions ask candidates on a computer to share their whole screen, recorded with the interview. It is off unless the position turns it on, and a candidate on a phone is never marked down for not sharing.
- One person, many names. If the same person applies under different names, Motium notices through one-way fingerprints of the email, phone, résumé text and connection, matched over 90 days. If the match is with another employer, that employer is never named.
What Motium does not do
- No eye or gaze tracking.
- No voice analysis for stress or honesty.
- No "AI-written answer" detector. These detectors are unreliable, and a 2023 Stanford study found they flag writing by non-native English speakers as AI far more often.
- No searching social media to build a profile.
A person decides what a signal means
Every signal reaches the reviewer with its time, above the recording. Nothing else reads it: not the scores, not Wren's review, not any ranking. A candidate who left the page for ten seconds might have been cheating, or their child walked in. The reviewer can jump to that moment and see. That is the whole point: the software notices, a person judges.
Candidates are told about these checks before the interview starts, on the same page that explains what they need. Telling people up front stops most cheating before it starts, and it is fairer than catching them out.
What this will not catch
A determined person with a second device out of camera view can still get help. No honest system catches everything without watching people in ways we will not do. The better defense is the interview itself: a live conversation with follow-up questions is much harder to fake than a one-way recording, because a rehearsed or fed answer falls apart at "tell me more about that." How the report presents these checks is in what a good interview report looks like. What candidates are told is in what to expect in an AI interview.
Questions people ask
Does Motium track eye movement?
No. Motium does not track eyes or analyze voices. It notes facts about the session, such as leaving the page or pasting text.
Does a fraud signal reject a candidate?
No. Signals reach the reviewer with their time. They never change a score, and a person decides what they mean.

