What Is Cluely AI and How to Block It During Online Exams

A candidate finishes a 90-minute certification exam in eleven minutes. The recording looks completely normal. No phone in view, no whispering, no second monitor. Weeks later, the credentialing body notices something odd: identical wrong-answer clusters showing up across candidates who have never met. That pattern is often the first real clue that Cluely was involved, not a phone hidden under the desk.

Cluely is an AI-powered desktop overlay that reads whatever is on a screen or listens in on a conversation, then feeds answers back to the user in real time and out of sight. It was built specifically for cheating, and its own founders have said as much publicly. For certification bodies, universities, and corporate hiring teams, that changes the entire proctoring conversation.

This article breaks down what Cluely actually does, why standard screen-share monitoring tends to miss it, and what a proctoring platform needs in place to catch it.

What Is Cluely, Exactly?

Cluely is a desktop application built by Columbia University students Roy Lee and Neel Shanmugam, who were suspended from the school in 2025 after creating an earlier tool called Interview Coder, designed to help engineers cheat on technical interviews undetected. Cluely is the commercial follow-up. According to TechCrunch, the company raised $15 million from Andreessen Horowitz in June 2025, on top of $5.3 million in earlier seed funding, putting total funding past $20 million.

The product itself is simple in concept and difficult to catch in practice. It sits on top of whatever application is open, whether that's a browser, a test portal, or a video call, as a transparent layer that's invisible to screen recording and screen-sharing tools. It can read on-screen text and quietly generate answers to exam questions, or listen through the microphone during a live interview and produce a spoken response the candidate reads back naturally.

It helps to think of it less like a chatbot and more like a teleprompter only the test-taker can see.

Why Standard Screen Monitoring Misses It

Three design choices make Cluely genuinely hard to catch with conventional tools.

• It bypasses keyboard and application logging. Cluely relies on hidden global keyboard shortcuts instead of visible menu clicks, so activity logs that only track standard application switches show nothing out of the ordinary.
• It hides from screen capture. The overlay renders in a way most screen-sharing and recording software simply doesn't pick up, so a recorded exam session can look entirely clean even while the overlay is active and generating answers.
• It fakes tab focus. The application tricks the browser into registering the exam tab as still active, so tab-switch detection, a common feature of basic exam-lockdown tools, never triggers.

None of this is theoretical. Adoption has moved quickly. Fabric, a company that builds AI interview-screening tools and tracks cheating patterns across its platform, analyzed 19,368 technical interviews between July 2025 and January 2026 and found cheating signals in 38.5% of candidates overall, rising to 48% in purely technical roles. Dedicated overlay tools, Cluely and Interview Coder among them, accounted for 45% of the flagged cheating methods. An earlier analysis of 50,000 candidates by the same company found unauthorized AI use had roughly doubled in six months, from 15% in June 2025 to 35% by December.

The risk isn't limited to individual dishonesty, either. The FBI's public alert on North Korean IT worker threats describes operatives using AI tools, including face-swapping technology, to pass video interviews and technical screens while posing as legitimate applicants under stolen identities. The United Nations estimates the broader scheme generates between $250 million and $600 million a year for North Korea, with operatives active in more than 40 countries. The FBI's own suggested defense is strikingly manual: ask the candidate to wave a hand in front of their face on camera, since doing so can break real-time AI face-rendering. What started as a shortcut for anxious candidates has become a documented state-level fraud vector.

Is Cluely Safe to Allow, Even If an Institution Wanted To?

Setting the cheating question aside, Cluely's own data practices carry real risk for any institution that might consider tolerating it. Per its published subprocessor list, current as of August 2025, the platform routes audio through Deepgram and AssemblyAI for transcription, and text through OpenAI, Grok, and Anthropic for response generation, across 15 named subprocessors. Cluely discloses who each vendor is but not exactly what each one can see or retain, so a credentialing body or university has no real way to verify where exam content ends up once the tool has "seen" it.

For a program protecting a proprietary exam bank, that's not a minor compliance footnote. It's a direct pipeline for questions to leak the moment a single test-taker runs the tool during a live exam.

What a Proctoring Platform Needs to Catch Cluely

Blocking an invisible overlay isn't a matter of smarter screen recording. Screen recording is exactly what Cluely is built to defeat. Catching it requires visibility that doesn't depend on what the screen feed shows.

A serious proctoring solution should provide, at minimum:

• Environmental monitoring independent of the screen feed. A second camera angle, or ideally a full-room view, can catch a candidate reading from a hidden overlay even when the screen recording shows nothing wrong.
• Monitoring of prohibited applications and browser activity that goes beyond simple tab-focus tracking, so hidden overlays and background processes get flagged rather than quietly overlooked.
• Behavioral analysis focused on the candidate, not just the screen. Gaze patterns that repeatedly drift to a fixed point off to the side, or answer timing that's implausibly fast and consistent, are signals a Cluely-style tool leaves behind even when the software itself stays hidden.
• Audio monitoring for live interviews and spoken assessments, since Cluely's interview mode works by listening, not by reading the screen.

This is where Proctor360's 360° Total View™ headset changes the equation. Because it captures a full view of the testing space independent of anything happening on the candidate's screen, an invisible overlay has nowhere left to hide. A proctor or an AI reviewer can see a candidate glance repeatedly at a fixed point on the desk, or notice a posture change that lines up with unnaturally fast, uniform answer timing, regardless of what the screen recording shows. Paired with the monitoring built into Proctor360's AI-Powered Auto Proctoring service, which flags prohibited applications and browser activity rather than relying on tab-focus alone, that closes the exact gap Cluely was designed to exploit.

For higher-risk exam content, that same environmental visibility is also available through Proctor360's government-focused remote proctoring solution, built for institutions that need CUI-grade handling, an option most proctoring vendors don't offer at all.

A Policy Note, Not Just a Technology One

Blocking software only gets an institution halfway there. Test-takers need to understand, before the exam starts, that this kind of AI assistance is a clear violation, not a gray area. Programs that pair plain, upfront disclosure with real environmental monitoring tend to see fewer disputes afterward, simply because the rule was stated clearly and the evidence, when flagged, isn't ambiguous.

Frequently Asked Questions

Is Cluely actually undetectable? Not to every method. It defeats screen recording and standard tab-focus tracking by design, but multi-angle environmental monitoring and behavioral analysis, things like gaze, timing, and posture, still catch its effects even when the software itself stays invisible.

Can Cluely be used during live video interviews, not just written exams? Yes. Its interview mode listens through the microphone, picks up the question, and generates a response the candidate can read back naturally within one to two seconds, according to reporting on how the underlying technology works.

Does Cluely collect data during use? Yes. Per its own subprocessor disclosure, it routes audio through transcription vendors, session and account data through cloud infrastructure providers, and generated responses through OpenAI, Grok, and Anthropic, across 15 named subprocessors, without detailing exactly what each one retains.

How is this different from a student using ChatGPT in a separate tab? A separate tab is visible to tab-switch detection and screen recording. Cluely's overlay is specifically built to avoid both, which is why catching it requires monitoring that doesn't rely solely on the screen feed.

How common is AI-assisted cheating in technical assessments right now? Fabric's analysis of 19,368 technical interviews found cheating signals in 38.5% of candidates overall, and 48% in purely technical roles, with dedicated overlay tools like Cluely responsible for 45% of the cheating methods detected.

What should a certification body do first if it suspects Cluely use in past exams? Start with a pattern review: identical wrong-answer clusters, implausibly uniform completion times, and gaze data if it was captured. A proctoring solution with environmental and behavioral monitoring built in can flag these patterns going forward instead of relying on manual review after the fact.

Talk to Proctor360 About Closing the Gap

If a current proctoring setup only watches the screen, an invisible overlay like Cluely was built to walk right past it. Reach out to Proctor360 to see how 360° Total View™ and AI-Powered Auto Proctoring monitor the candidate and the room, not just the browser tab.

Sources: TechCrunch reporting on Cluely's June 2025 funding round; Fabric's analysis of AI interview cheating trends (2026); the FBI's public alert on North Korean IT worker threats; Cluely's published subprocessor disclosure (current as of August 2025).