TL;DR. Cheating in an interview with AI is not a single thing. Cluely raised $15M from a16z in June 2025 (TechCrunch), yet Amazon rescinded the offer and Columbia suspended Roy Lee. And AI Act Article 50 becomes applicable on 2 December 2027 (European Commission). Three zones, one clear line — the one most articles refuse to draw.
A Columbia student lands an Amazon internship with an invisible AI overlay. He posts the video on YouTube. Amazon pulls the offer. Three months later, he raises $15M.
You're prepping your own interview with ChatGPT tonight. At what point do you cross into cheating?
No moralizing here. A three-zone taxonomy, a decision tree, and Velyq's stated position.
"Cheating with AI in interviews": why the question is framed wrong in 2026
Most articles blur two very different things: being coached by an AI before the interview and letting an AI sit the interview for you. That's like confusing sparring practice with a fixed fight. Velyq draws the line.
The live-cheating market is not marginal. Roy Lee posted on X that Cluely hit $228,500 in revenue in month 2 (X / Roy Lee). Andreessen Horowitz put $15M on the table at a $120M post-money valuation (TechCrunch). The startup advertises engineering salaries between $300K and $1M per year (Cluely careers).
On the community side, one widely shared Hacker News thread captures the mood: "The engineering interview process is broken, AI cheating is exposing it faster" (HN search). AI cheating is less a cause than a symptom of a broken interview format.
But "the format is broken" tells you nothing about what you should do. Before you decide, you need precise vocabulary.
March 2025 — Roy Lee, a Columbia sophomore, posts the YouTube video of his successful Amazon interview using Interview Coder. Amazon rescinds the offer, Columbia suspends him for a year. June 2025 — Cluely (formerly Interview Coder) raises $15M from a16z at a $120M post-money valuation. March 2026 — Roy Lee admits he publicly lied about Cluely's revenue numbers in 2025 (TechCrunch, March 2026).
The Velyq taxonomy: 3 zones to separate prep, live assist and delegation
Three uses of AI in interviews, three legal regimes, three career-risk levels. Not mixing them is half the work.
Zone 1 — Legitimate prep. You train upstream on a simulator (Velyq, ChatGPT in coach mode, a peer mock). It's asynchronous, outside the real interview, and you show up alone on the day. No candidate contract forbids it. No law blocks it. It's the 2026 equivalent of Cracking the Coding Interview.
Zone 2 — Live cheating. You run Cluely, Interview Coder or an equivalent during the real interview. The overlay reads the question via screen capture, generates an answer, whispers it to you. Invisible to recruiters today — not tomorrow. And almost always a breach of the candidate agreement.
Zone 3 — Full delegation to an AI agent. Video avatar, voice clone, agent answering on your behalf. You're no longer in the interview — a system is. Directly in scope for AI Act Article 50 (transparency) and GDPR Article 22 (automated decision-making).
- ✓Legality: fully legal
- ✓Detectability: N/A
- ✓Learning value: high, durable
- ✓Career risk: zero
- ✓Velyq position: yes — this is our job
- ✗Legality: contract breach (Z2) / AI Act 50 + GDPR 22 (Z3)
- ✗Detectability: window closing fast
- ✗Learning value: zero — renting a crutch
- ✗Career risk: rescinded offer, blacklist, probation
- ✗Velyq position: no — we flag it
What contracts and law actually say (Amazon, AI Act, GDPR)
Three texts already frame AI use in interviews, and the "it's just a small assist" argument fails all three.
The candidate contract. After the Roy Lee story, an Amazon spokesperson reminded that candidates "agree not to use unauthorized tools during the process" (Gizmodo). The same is true at Meta, TikTok, and most large US and European employers. Live cheating is not a fuzzy moral lapse — it's a documented contractual breach.
AI Act Article 50. It requires transparency on any AI-generated or AI-modified content. If you show up via a video avatar or a voice clone, you must disclose it. The high-risk employment rules (Article 6 + Annex III, covering AI tools used for recruitment and candidate evaluation) become applicable on 2 December 2027 (European Commission).
GDPR Article 22. It already prohibits a fully automated hiring decision without human review (full text). By symmetry, delegating 100% of the interview to an AI agent pushes the candidate out of the legal evaluation frame: it's no longer you being assessed, it's your agent.
Career risk 2025–2026: Amazon, Columbia and the detection market
The "nobody will see it" argument has already been falsified — at several levels.
Amazon rescinded Roy Lee's internship offer after his YouTube video went up (Business Insider). HR precedent on the employer side: you can be unmasked after the offer, after the signature, and the offer evaporates.
Columbia suspended Roy Lee for a full year, along with his co-founder Neel Shanmugam (Business Insider). The academic side also sanctions — and the record follows the candidate.
On the detection side, the counter-market is arming fast. DeepCue / Anti-Cluely detects virtual machines, invisible windows and abnormal gaze patterns (anti-cluely.com). ATS vendors are shipping their own modules. The invisibility window is closing fast.
Even the movement's poster child has lost credibility: in March 2026 Roy Lee admitted he had publicly lied about Cluely's 2025 revenue (TechCrunch). The "$228K MRR in month 2" story that triggered the a16z round was inflated.
Candidate decision tree: 4 checks before using AI in an interview
You're staring at an AI tool ahead of your next interview. Ask yourself these four questions in order — a single no pushes you out of the safe zone.
1. Legal. Does the use violate the candidate contract you sign upstream? From December 2027, does it violate AI Act Article 50? If yes, stop.
2. Ethical. Would you be comfortable if your recruiter could see your full screen live? If that thought makes you flinch, you have your answer.
3. Career risk. If the offer is rescinded like Roy Lee's, or if your reputation is burned on LinkedIn, can you absorb it? For most profiles, the answer is no.
4. Real learning value. Are you improving for the next interview, or renting a crutch at $60/month (Business Insider)? A rented crutch has never built muscle.
Velyq's position, stated and public: we build the Zone 1 tool (prep, simulation, async feedback), we refuse to build Zone 2, and we flag Zone 3.
Frequently asked questions
Is using ChatGPT to prep STAR answers cheating?
No. That's Zone 1 (asynchronous prep). No candidate contract forbids it. It's the equivalent of an HR coach or a prep book.
Is Cluely legal in Europe in 2026?
The tool is legal to buy, but its use in an interview almost always breaches the candidate contract. From 2 December 2027, AI Act Article 50 will also require AI-content transparency.
Can recruiters really detect an AI overlay?
Yes, more and more. DeepCue, Anti-Cluely and ATS vendors embed detectors for virtual machines and gaze patterns. Dedicated article coming.
If I cheat and nobody sees it, where's the problem?
Three problems: (1) you sign for a role you're not calibrated for, (2) probation exposes you brutally, (3) Roy Lee was outed by a third party six months later — impunity doesn't last.
What if I disclose that I'm using AI during the interview?
That's compliant with AI Act Article 50, but it changes the test: the recruiter now evaluates your prompting, not your domain skills. A few companies accept it (rare), most don't.
Why are so many candidates cheating then?
Because the format is broken (a recurring HN signal) and because tools are cheap ($60/month). But "everyone does it" is neither a legal nor a strategic argument.
Is Velyq a cheating tool?
No. Velyq is Zone 1: asynchronous simulation, structured feedback, training before the interview. On the day, you face the recruiter without a live assist.
Key takeaways
- Three zones, not two: prep, live cheating, full delegation — don't mix them.
- Cluely raised $15M, but Amazon pulled the offer and Columbia suspended — the career risk is real.
- AI Act Art. 50 + GDPR Art. 22 already frame Zone 3; the frame tightens on 2 December 2027.
- The detection market (DeepCue, Anti-Cluely) makes Zone 2 increasingly disposable.
- Velyq's position: prep yes, live crutch no.
- Before any AI use: run the 4 checks (legal, ethical, career, learning value).
Two concrete actions now:
- Train in Zone 1 on the Velyq AI interview platform — async simulator, zero live assist.
- Audit your CV before the interview via the Velyq CV review — the best prevention is still a solid file.
Read next
- Detecting AI in interviews: how recruiters spot an overlay in 2026
- Free AI interview simulator: ChatGPT, Hellowork, Huru or Velyq
- The AI Act and recruitment: what changes on 2 December 2027 for HR
- Behavioral interview questions: the STAR method without an AI crutch
- Probation and termination: why interview cheating breaks later
