Cheating Your Interview With ChatGPT or Cluely: How Recruiters Detect You in 2026

TL;DR. Between June and December 2025, the share of candidates cheating in interviews with an AI jumped from 15% to 35% (Fabric study, 19,368 interviews analyzed). On the other side, recruiters have equipped their processes with detection: eye tracking, latency analysis, dual-voice, platforms like Fabric, Sherlock, or HackerRank Proctor that cross-reference 20+ signals. Amazon has already pulled offers retroactively, Anthropic rewrote its questions. Cheating still works, but the window is closing fast — and the consequences, they're durable.

We're not here to moralize. We're here to map what's actually happening: who cheats, how, with which tools, how recruiters detect, what you risk when caught, and why the economic question of "cheat or not" no longer has the same answer as 12 months ago.

The state of play: interview cheating has become an industry

In 18 months, an entire ecosystem has formed around the idea "let's take interviews for you."

The main tools today:

• Cluely (formerly Interview Coder). Raised $5.3M in April 2025, founded by Roy Lee (expelled from Columbia after the Amazon incident). Invisible overlay that reads your screen and mic, whispers the answer in 1-2 seconds. Founder's claimed slogan: "cheat on everything". TechCrunch
• LockedIn AI, Parakeet AI, Final Round AI (former simulator turned live copilot), Interview Coder (integrated into Cluely)
• Typical business model: $20-50 / month. Against a $150,000 engineer's annual salary, the equation seduces — and that's exactly the risk/reward ratio that Fabric highlights in its February 2026 report.

Why recruiters have caught up (fast)

The argument you still heard in early 2025 — "HR won't see a thing" — no longer holds. Three things happened.

1. Technical test platforms doubled their detection capacity

HackerRank publishes an integrity score (High / Medium / Low) on every test, with session replay, screenshots, code pattern analysis. Their AI plagiarism system reaches 93% accuracy, 3× more accurate than classical methods. CodeSignal now flags 35% of proctored assessments (40% on entry-level roles), with a multi-layered "Suspicion Score."

2. Startups specialized in detection emerged

Fabric analyzes 20+ live signals (gaze, response timing, linguistic analysis) and produces a timestamped cheating-probability score with evidence. Sherlock AI and InCruiter offer similar solutions for live interviews. When InCruiter launched its deepfake detection in early 2026, the platform identified 25-30% fraud on flagged sessions — nearly double what experienced human recruiters detected before.

3. Big companies wrote their own rules

• Amazon pulled Roy Lee's offer after seeing his Interview Coder YouTube video, published an internal guide for its recruiters with alert signals, and officially banned generative AI use in live interviews.
• Anthropic had to rewrite all its technical interview questions in 2026 because candidates were using… Claude (their own product) to answer them. Their official policy now permits AI for interview prep but not during the live, unless instructed otherwise.
• 62% of tech organizations explicitly ban AI in technical interviews. Karat reminds us that live interviews remain the best defense because "it is extremely difficult to simulate the reasoning of a dev who copy-pastes AI output."

The 8 signals recruiters watch in 2026

Here's what Fabric, Sherlock, HackerRank, and detection-trained recruiters observe concretely. If you're applying today, assume this grid is running in the background of your interview.

1. Eyes that "read" instead of looking at the camera

When you read an AI response, your eyes follow a left-right, line-by-line scanning pattern. The camera sees that. Eye tracking algorithms (used by InCruiter, Sherlock, or baked into proctoring platforms) flag candidates who "look at an invisible teleprompter" rather than engage with the interviewer.

Human corollary: an experienced recruiter notices that you look "down-left" on every answer — where the Cluely overlay displays.

2. The uniform 3-5 second latency

A genuine candidate responds with variable latency: 0 seconds on "how old are you?", 10 seconds on "tell me about a conflict you handled." A candidate waiting for an AI: systematically 3-5 seconds of silence before every answer, even on simple questions. That's the time the LLM takes to generate. The pattern is too regular to be natural.

3. Signature AI phrasing

LLMs have tics: "Certainly, let me walk you through this…", "Here's a comprehensive approach…", "It's important to note that…", always-balanced 3-point lists, wrap-up conclusions. Technical recruiters know these patterns by heart from seeing them daily in code reviews.

4. Mechanical question repetition

"So if I rephrase your question correctly, you're asking me…" — once or twice in an interview is normal. On every technical question is a stall tactic to let the AI process the request. Fabric and Sherlock auto-flag this behavior.

5. Dual-voice in the background

Modern platforms detect the presence of a second voice or faint synthesized speech noise. Some candidates using earphones with TTS sometimes let sound leak into their ambient mic. Testlify, InCruiter, HackerRank listen for that.

6. Tab-switching and window activity

On technical assessments (HackerRank, CodeSignal, Codility), every tab switch, window open, or full-screen exit is logged. Session replay is viewable by recruiters after the fact. The "Suspicion Score" climbs, the recruitment stops.

7. Follow-up questions that go off-script

A well-formed AI response does not survive "okay, and if we changed constraint X, how would you adapt?". The classic trap for tech recruiters today: let the candidate give the "perfect" answer, then dig with 3-4 unforeseen follow-ups. The candidate who copied an AI response collapses — not the one who thought.

8. Pro tools that cross-reference all this in real time

What you risk when caught (and it happens)

The textbook case is Roy Lee. A Columbia student, he landed a software engineer offer at Amazon using Interview Coder. He filmed the operation. He posted the video on YouTube. Two days later, someone (never identified) sent the video to Columbia AND to Amazon.

Result:

• Amazon pulled the offer
• Columbia opened disciplinary proceedings — he was expelled
• Amazon flagged his ATS file (which, per Gizmodo, means he'll never reapply there)
• The virality of the story burned him across Silicon Valley — he bounced back by creating Cluely, but that's a pragmatic pivot, not a plan A

This is not an isolated case. The consequences we see on the company side:

And the scenario we forget: if cheating works and you're hired, the first 3-6 months expose the impostor anyway. A dev who had Cluely answer their algo questions can't code. Peers see it. PRs don't pass. 1:1s get awkward. Probation termination isn't a bug, it's the feature of probation.

And in France? The paradox of a country stuck between ethics and tolerance

The FR landscape has two speeds. On one side, the APEC positions AI as an ally — preparation, CV/cover-letter writing, simulation — and signals that 15% of executives have already used AI to draft their CV or cover letter. It's considered a normal evolution of practices, not cheating.

On the other, Slate.fr documents the explosion of "how to cheat in an interview with ChatGPT" TikTok tutorials, and Hellowork headlines "AI cheating tools are exploding, here's why you should avoid them" — meaning even media historically friendly to AI in recruiting now clearly draw the line.

The FR norm in 2026 is taking shape:

• ✅ Preparing an interview with AI (company research, question simulation, pitch rewording): accepted, encouraged by APEC
• ✅ Improving a CV/cover letter with AI (after a personal first draft): accepted
• ❌ Reading a generated answer in real time during the interview: cheating, sanctioned
• ❌ Impersonating someone else via deepfake or proxy: fraud, potentially criminal

The regulatory context pushes in the same direction: the EU AI Act enters progressive application starting August 2026, with increased transparency obligations on AI tools used in recruiting — on the company side, but the same logic will infuse the candidate side.

"So how do I train, concretely, without cheating?"

If you made it this far, you're probably not looking to cheat — you're wondering how to prepare the old-fashioned way, in 2026, without feeling stupid against someone who has Cluely.

The honest answer: working on your prep has become a competitive advantage again. Cheating tools are visible, interviews are getting smarter (unforeseen follow-ups, off-CV questions), and the only thing that really holds up is having already thought through the questions and already spoken out loud.

Velyq is part of these prep tools (that's why we write this article — we want to be transparent). We don't whisper answers in real time, and we never will. We make you rehearse, give you structured feedback on what you said, and leave you facing the camera when it counts. It's slow and requires work. It's also what holds up in 2026.

FAQ

Can ChatGPT really help me answer during an interview?

Technically yes, practically less and less. If you use ChatGPT in another tab during a video interview, tab-switching is detectable by assessment platforms. If you use a Cluely-type overlay, there are signals (uniform latency, scanning gaze) that trained recruiters catch. The real issue is that even if it passes the interview, it doesn't pass probation.

How do recruiters know I'm using Cluely or Interview Coder?

Multiple ways in 2026: eye tracking that spots a reading gaze, too-uniform response latency (3-5 sec before each answer), signature LLM phrasing, systematic question repetition, unforeseen follow-ups you can't answer. Platforms like Fabric or Sherlock combine 20+ signals and give the recruiter a real-time probability score.

Is it legal to use AI during a job interview?

It's not criminally illegal (except proxy or deepfake cases which can amount to identity fraud). But it's almost systematically a breach of pre-contractual good faith: the offer can be pulled, and if hire occurred, it can ground a termination for misconduct. Amazon, Anthropic, and most large tech groups explicitly ban AI use in live interviews — it's in the interview rules you agree to before starting.

What's the actual risk if caught?

Offer pulled (Roy Lee / Amazon case), company ATS blacklist (impossible to reapply), probationary termination if cheating is discovered after hire, school notification in university cases. In circles where everyone talks (tech, finance, consulting FR), reputation sticks — a recruiter remembers.

Can I use AI to PREPARE my interview without it being cheating?

Yes, largely. The APEC itself encourages this use: preparing answers to common questions, simulating an interview, reviewing your pitch, anticipating trap questions. As long as you don't use AI in real time during the live, you're in the zone accepted by every company. The practical rule: if you can tell the recruiter ("I trained with a simulator"), it's preparation; if you have to hide it, it's cheating.

Does AI cheating still work in 2026?

Partially, and less and less. On a classic HR video interview, with a recruiter not trained on detection, a tool like Cluely can still pass. On a technical interview on HackerRank, CodeSignal, or with a platform running Fabric/Sherlock behind, detection probability is high — and climbs every quarter. In 2025, 35% of proctored assessments were flagged for suspicion. In 2026, that share grows. The window is closing fast.

5 takeaways

1. AI interview cheating is no longer a rare case — 35% of tech candidates have used it as of December 2025. It's a market, with products, funding, a price tag (~$20-50/month).
2. Detection caught up with offense in 18 months. Eye tracking, latency, dual-voice, platforms like Fabric, Sherlock, HackerRank Proctor — recruiters no longer improvise.
3. Consequences are durable: offer pulled, ATS blacklist, reputation damaged in a sector where everyone talks. The Roy Lee/Amazon case is the textbook.
4. Even if cheating passes the interview, it fails probation. The first 3-6 months always expose the gap between the "perfect" candidate and the real dev.
5. Prep AI (not cheat AI) is a legitimate practice — APEC, Anthropic, 15% of FR executives. That's where the serious candidate's competitive edge plays out now.

Also worth reading

• Google Interview Warmup Alternative: Replacing the Tool Google Shut Down in 2026
• The 10 Most Asked Interview Questions in 2026
• Can AI Actually Help You Ace Your Interviews? The Honest 2026 Guide