How Candidates Cheat With AI in Technical Interviews (And How to Catch It)

Remote technical interviews have a new problem: a candidate can now have an AI copilot solve the exact question you just asked, in real time, without you seeing a thing. Here's exactly how it happens — and an honest look at what actually catches it.

The five ways candidates cheat with AI

AI-assisted interview cheating isn't one thing — it's a spectrum, from casual copy-paste to purpose-built invisible software. Understanding the methods is the first step to designing interviews that resist them.

1. Copy-paste from a chatbot

The simplest method: the candidate reads your question, types or pastes it into ChatGPT or Claude in another window, and pastes the answer back. It's common because it's effortless — and it's also the easiest to detect, because a large block of text appearing with no corresponding typing is a strong tell.

2. A second monitor or device

Instead of switching windows, the candidate keeps the AI open on a second screen, a tablet, or a phone propped against their monitor. They read the answer while appearing to look at their main screen. This defeats anything that only watches the interview window.

3. Invisible overlay tools

A new class of purpose-built apps — Cluely, Interview Coder, LeetCode Wizard and similar — screenshot the coding question, send it to an LLM, and display the answer as a floating overlay that is deliberately excluded from screen capture. They're designed so the answer is invisible to both a screen share and a screenshot. This is the most sophisticated on-device method.

4. Live voice copilots

Tools that listen to the interviewer's audio and generate spoken or on-screen answers in near real time. Paired with a phone or a second device, they turn a behavioral interview into a read-aloud exercise.

5. A person off-camera

The lowest-tech and hardest-to-detect method: someone out of frame feeding answers, or the candidate simply reciting memorized solutions. No software is involved at all.

What each method leaves behind

Every method except the last two leaves a behavioral or environmental fingerprint. This is what integrity monitoring is built to surface:

• Copy-paste: large paste events with little or no proportional typing.
• Window switching: the interview tab losing focus, being hidden, or backgrounded — with timing that correlates to when a question was asked.
• Second monitor: the browser can detect that more than one display is attached.
• Overlay tools: a required full-screen share makes anything rendered into captured pixels visible; a native lockdown client can enumerate the running processes and match them against a database of known cheating apps.
• Reading a second device: a webcam gaze check can flag sustained looking-down or looking-away, and requiring earphones prevents a nearby device from hearing the interview audio.

How to actually catch it

A layered approach works far better than any single check:

1. Require a full-screen share (not a tab). This alone forces overlay tools to either reveal themselves or exclude themselves — and the exclusion itself is suspicious.
2. Monitor behavioral signals — paste bursts, focus loss, tab hiding — and surface them live to the interviewer.
3. Require the camera on, and if you can, a webcam presence check that flags repeated looking-away toward a second device.
4. Require earphones so a nearby phone can't hear the interview audio to answer it.
5. Ask adaptive follow-up questions. This is the single most reliable defense — see below.

InterviewGuard is built around exactly this layered model: a candidate opens a link (no install), consents, and shares their screen, while the interviewer watches a live risk score and the underlying signals. See how it works.

What no tool can catch (and why honesty matters)

Any vendor claiming to be "100% AI-proof" is selling you a false sense of security. Be clear-eyed about the limits:

• A second phone or laptop off to the side is invisible to any software running in the browser — or even on the candidate's computer. A gaze check helps, but can't see the device itself.
• A person off-camera feeding answers leaves no digital trace.
• Memorized solutions are indistinguishable from genuine knowledge to any automated system.

The only reliable defense against these is process, not software: a second "room camera" showing the desk, and — most importantly — live, adaptive questioning.

A practical interviewer checklist

• Require entire-screen sharing and camera-on before the interview starts.
• Ask the candidate to explain their code line by line, then change a requirement mid-solution ("now handle duplicates," "now make it thread-safe"). An AI-fed candidate stalls; a real one adapts.
• Watch for answers that arrive fully-formed and optimal after a long pause.
• Use a live risk score as a triage aid — investigate flags with questions, don't auto-reject.
• For high-stakes rounds, add a second-angle camera or an in-person final.

Related reading: Cluely and Interview Coder: how to detect them · Is remote interview proctoring legal?

InterviewGuard provides consent-based behavioral integrity monitoring. It is a triage aid, not proof of misconduct, and should always be paired with human judgment and follow-up questioning.