How to Prevent Cheating in Remote Interviews

A candidate gives a clean, well-structured answer to your system design question. You ask one follow-up: why put a queue there instead of calling the service directly? The answer stalls, restarts, and drifts back to generalities. Nothing on camera looked wrong. The problem appeared only when the conversation left the script.

That is what remote interview cheating often looks like now. It is rarely a crib sheet taped under the webcam. It is a second person, a second screen, or a model listening in and drafting answers faster than anyone could type them. Software cannot stop all of it, and monitoring that tries to catch everything ends up flagging honest people for ordinary behavior.

So the answer to how to prevent cheating in remote interviews is less about surveillance than about four things: interviews where outside help does not produce a good answer, identity checked before the first question, a few clear rules that candidates hear in advance, and a person who weighs the evidence.

What cheating in remote interviews looks like now

The common forms:

• A different person on camera. In proxy interviews, a more qualified stand-in takes the interview under the candidate's name, or answers off camera while the candidate lip-syncs. In 2022 the FBI's Internet Crime Complaint Center reported a rise in complaints about people using deepfakes and stolen personal information to apply for remote jobs.
• Someone off screen feeding answers. A friend in the room, a chat window on a phone, a voice in an earbud.
• AI-generated answers read aloud. Tools can transcribe the interviewer's question and put a suggested answer on the candidate's screen within seconds, sometimes in a window that does not show up when the screen is shared. This is what most people mean by AI interview cheating, and it is hard to spot by watching alone.
• Copied code. A solution pasted from a model or a practice site, or typed from memory without being understood.
• A different person in later rounds. The strong candidate from round one is not the one in round three, or on the job.

Design interviews where cheating does not help

The strongest defense is not detection. It is design.

Anchor questions in the candidate's own experience. "What's the hardest bug you've fixed?" invites a generic story. "Walk me through the last production incident you handled. What did you check first, and what did you rule out?" asks for details only someone who was there would have, and every answer opens another question. A real story gets more detailed under questioning. A borrowed one gets vaguer.

Use work samples. A short task close to the real job, done live, tells you more than a description of past work. Watch how the work gets done, not only the result.

Follow up on what the candidate just said or wrote. It is the most useful habit of all, and the questions are simple:

• "Why this approach rather than the obvious alternative?"
• "What would you change if the input were a thousand times larger?"
• "Extend this to handle refunds."
• "You said the team pushed back. What was their argument?"

Prepared answers, whether a friend wrote them or a model did, are prepared for the question. They are not prepared for a question about the answer. Each follow-up depends on what the candidate just said, so any helper, human or software, has to catch up on every turn.

Be careful about what you conclude. A pause on its own proves nothing. And you do not need to prove cheating to score it. An answer the candidate cannot explain, or code they cannot walk through, earns a low score on a good rubric whatever the reason. That is also the fairest outcome: you score what the candidate showed, not what you suspect.

Keep it structured. Write the questions, the follow-ups and the scoring guide before the first candidate, and use the same ones for everyone. Structured interviews predict job performance about twice as well as unstructured ones (Sackett et al., 2022), and a written guide makes it plain when an answer sounded polished but did not show the skill. Our structured interview rubric template is a good place to start.

Verify identity before the first question

A proxy is cheapest to stop at the door. Check a government ID against a live selfie before the interview starts, not afterward, when the interview time is already spent. Announcing the check in advance also discourages the attempt.

Then keep the same person across rounds:

• Have later-round interviewers look at the verified selfie, or a minute of the earlier recording, before they begin.
• Build later questions on earlier answers. "Last time you said you would split the service by customer. What changes if one customer is half your traffic?" A stand-in will not know what was said.
• Have whoever runs onboarding confirm that the person who starts the job is the person you interviewed.

Set the rules up front

Before the interview, tell candidates in plain language:

• what is allowed: notes, documentation, a calculator, AI tools, or none of these
• what is monitored, such as switching tabs or leaving full screen
• what happens if a rule is broken
• why you do it

Part of this is fairness. Nobody should lose a job over a rule they did not know existed. Part is consent: people deserve to know how they will be evaluated before they agree to it. And it cuts false alarms. A candidate who knows that leaving full screen counts is much less likely to do it by reflex to dismiss a notification.

Notice can also be a legal requirement. Illinois's Artificial Intelligence Video Interview Act, for one, requires employers that ask applicants for Illinois-based jobs to record video interviews, and use AI to analyze them, to notify candidates, explain how the AI works and get their consent first.

Monitor signals that are clear and fair

Switching to another tab, leaving full screen and taking a screenshot are clear, discrete events. Each has one plain meaning and a timestamp, a candidate who knows the rule can avoid it, and a reviewer can check the recording at that moment. None of them catches everything. That is fine. Their job is to enforce rules the candidate already knows, not to read minds.

Other signals sound rigorous but are not fair. Gaze or eye tracking is the clearest example. People look away to think. They glance at notes they are allowed to have. Glasses reflect the screen. A candidate whose camera sits on a second monitor looks "away" for the whole interview. Someone reading the question on screen looks a lot like someone reading an answer. Some neurodivergent candidates avoid eye contact or look away to concentrate. Each puts a flag on an honest person that a reviewer then has to disprove. The same goes for pauses, since people pause to think or to translate, and for pop-ups, which are the operating system's doing, not the candidate's.

Nothing in Hyrr watches a candidate's eyes during the interview, and there is no voice-biometric check. The review of the recording afterwards may note where a candidate looked, but that note is never scored, never counts as a strike and never ends an interview.

Make interview integrity review human and evidence-based

A flag is a question, not a verdict. Sound integrity review has three parts:

1. Evidence tied to the moment. Every finding should say in plain language what happened and link to that point in the recording, so a reviewer can watch it instead of trusting a label.
2. Automatic consequences only for clear, announced rules. Ending an interview after a candidate breaks a known rule twice is defensible. Rejecting someone because software decided their eyes moved oddly is not.
3. A person decides. Ambiguous cases go to someone who watches the moment and weighs it against the whole interview. If it is still unclear, ask the candidate, or offer a second conversation.

Keep a short record of what was flagged, what the reviewer saw and what was decided, in case the decision is ever questioned.

Decide your AI policy explicitly

Candidates using AI in interviews are not automatically cheating. It depends on what you told them. There are three coherent policies:

The worst policy is an unstated one. Some candidates assume AI is fine, others assume it is forbidden, and you end up comparing people who took different tests. For the third option, see our guide to assessing AI fluency. Hyrr's Practical AI Fluency assessments, which are new and rolling out, take that route, with in-interview AI exercises and hands-on labs that measure how well a candidate works with AI tools.

A checklist for your next remote interview

• Before the first candidate, write the questions, the scoring guide and follow-ups that depend on the answer: why, what if, now extend it.
• Anchor some questions in the candidate's own work and experience.
• Include a live work sample, and ask about it while it happens.
• Check a government ID against a live selfie before the first question.
• Link later rounds to earlier ones with the verified selfie and questions that build on earlier answers.
• Tell candidates in writing, before they start, what is allowed (AI included), what is monitored and why.
• Strike only on clear, avoidable events, never on gaze, pauses or pop-ups.
• Link every flag to the moment in the recording, and have a person review it.

How Hyrr handles interview integrity

Hyrr runs the first interview round as a live AI interview. Here is what it does on integrity:

• Identity checked first. On roles where it is switched on, candidates complete a government ID and selfie check through Stripe Identity before the first question.
• Strikes only for clear events. On monitored roles, switching tabs, leaving full screen and taking screenshots count as strikes, and two strikes end the interview. Looking away, pop-ups and pauses never count.
• Follow-ups on the candidate's own answers. Your team writes follow-ups for each question, and the AI interviewer generates its own from what the candidate says.
• Hands-on work, seen as it happens. In the shared code editor and whiteboard, the AI sees the code as it is typed and the diagram as it is drawn, and asks the candidate about it. The work is saved to the report.
• A review you can check. Each interview gets one plain-language integrity review, linked to the moment in the recording.
• A person makes the call. Hyrr does not make the hiring decision. Its scores inform a decision your team makes.

For more on the checks themselves, see Hyrr's identity and integrity stack. To see the follow-ups from the candidate's side, pick a sample job in the live demo on our home page and take a real AI interview in your browser in minutes, or try the frontend engineer demo.