Cluely raised $5.3 million by promising candidates they could cheat on everything. Their $75/month tier specifically advertises complete invisibility during screen sharing. The pitch is simple: an AI reads your screen, listens to the interviewer, and feeds you perfect answers through an overlay that nobody else can see.
We wanted to know if that claim held up against an AI interview platform built to catch exactly this kind of thing.
So we bought a Cluely subscription, set it up on a test machine, and ran it against a Fabric interview. This blog walks through what we did, what happened, and what the Fabric report actually showed.
Why did we test Cluely against Fabric?
Cluely has become the most talked-about interview cheating tool of the past year. Founded by two former Columbia students who were expelled for using an earlier version of the tool to land internships at Amazon, the company has leaned into controversy as a growth strategy. Their manifesto literally tells people to start cheating.
For hiring teams, Cluely represents a specific and growing problem. The tool uses low-level graphics hooks to render its overlay directly on the GPU output, bypassing screen-sharing software entirely. Traditional proctoring methods that monitor tab switches or flag second screens cannot see it.
We had already built detection capabilities into Fabric's interview platform. But claims are easy to make. We wanted to put our own system to the test with the most aggressive cheating tool on the market, document the entire process, and share the results publicly.
How did we set up the experiment?
We kept the setup simple to mirror what an actual candidate using Cluely would do.
1. Purchased Cluely's top tier
We subscribed to the $75/month plan, which Cluely markets as their undetectability tier. This plan includes everything in their Pro subscription plus what they describe as complete invisibility during screen sharing.
2. Configured the cheating tool
We installed Cluely on a test machine, enabled audio capture and screen reading, and confirmed the overlay was invisible when sharing the screen via video conferencing software. The tool was fully operational and feeding real-time AI-generated responses.
3. Ran a standard Fabric interview
We initiated a Fabric interview for a technical role and answered every question using the responses Cluely generated. We did not supplement Cluely's answers with our own knowledge. If Cluely provided it, we said it.
The goal was to give Cluely every possible advantage. Top-tier plan, proper setup, and complete reliance on its outputs.
What happened during the interview?
Cluely worked as advertised on the surface. It captured the interviewer's questions through audio, generated responses within a few seconds, and displayed them on the invisible overlay. The answers were detailed, structured, and technically relevant.
From the candidate's perspective, the experience felt seamless. Questions came in, answers appeared, and the conversation moved forward.
But Fabric was watching signals that Cluely could not mask.
How did Fabric catch the cheating?
Fabric flagged the interview with an 85% cheating probability. The detection report broke this down across three signal categories.
1. Response timing consistency
Every answer followed an almost identical timing pattern. Regardless of whether the question was simple or complex, there was a consistent delay of a few seconds before the response began.
In a genuine interview, a candidate answers a question about their background almost instantly but takes longer to work through a system design problem. Cluely flattened that natural variation into a predictable rhythm, and Fabric's timing analysis caught it.
2. Language structure and AI-typical phrasing
The answers had a uniformity that real human speech rarely produces. Responses consistently followed structured patterns, used similar transitional phrases, and maintained a level of grammatical precision that is characteristic of LLM-generated text.
Fabric's content integrity engine compares response patterns against known LLM output characteristics. The Cluely-generated answers triggered multiple flags for AI-typical language construction.
3. Fabric's proprietary detection algorithm
Beyond individual signals, Fabric's detection engine combines over 20 distinct data points into a composite probability score. This includes behavioral patterns, interaction telemetry, and content analysis that work together to identify synthetic assistance.
The algorithm does not rely on any single signal. Even if a cheating tool manages to mask one category of signals, the combination of remaining indicators still produces a reliable detection. In this test, the convergence of timing, language, and behavioral data produced the 85% probability score.
What does this mean for companies hiring today?
This experiment confirmed something that matters for every hiring team dealing with high-volume technical recruiting: tools that claim to be undetectable can still be caught when the detection approach looks at the right signals.
Cluely's invisible overlay technology is sophisticated. It defeats screen recording, screen sharing, and traditional browser-based proctoring. Any company relying solely on those methods has a real vulnerability.
But the cheating tool cannot control how a candidate's behavior changes when reading AI-generated text instead of speaking from genuine knowledge. It cannot eliminate the processing delay inherent in its audio-to-text-to-LLM pipeline. And it cannot make AI-generated language sound like natural human speech.
Fabric's approach works because it does not try to catch the tool itself. It catches the behavioral footprint the tool leaves behind.
Conclusion
We spent $75 on Cluely's best plan and used it to cheat in a Fabric interview. Fabric caught it with 85% cheating probability and produced a detailed report showing exactly which signals triggered the detection.
The full video of the experiment and the Fabric report are linked above. Hiring teams can review both to see exactly what the detection looks like in practice.
If your current interview process cannot detect tools like Cluely, it is worth evaluating what signals you are actually monitoring. The gap between what traditional proctoring catches and what modern cheating tools can bypass is growing fast.
FAQ
Can Cluely actually be detected during screen sharing?
Cluely's overlay cannot be seen through screen-sharing software. However, the behavioral patterns it creates in candidates, such as consistent response timing and AI-structured language, can be detected by platforms analyzing those signals.
How much does Cluely cost?
Cluely offers a free tier with limited features, a $20/month Pro plan with unlimited AI responses, and a $75/month tier that adds undetectability during screen sharing. The $75 tier is specifically marketed for interview and meeting use.
What is Fabric?
Fabric is an AI interview platform that conducts conversational interviews and analyzes over 20 behavioral, interaction, and content signals to detect candidates using AI assistance tools. It provides detailed timestamped reports with cheating probability scores.
How accurate is Fabric at detecting cheating tools?
Based on extensive testing, Fabric detects cheating in 91% of cases accurately. The platform provides detailed reports showing which specific signals triggered the detection, allowing hiring teams to verify results independently.
Does Fabric work against other cheating tools besides Cluely?
Fabric's detection is tool-agnostic. Because it analyzes behavioral patterns and content signals rather than trying to identify specific software, it works against any AI assistance tool that creates similar patterns, including Interview Coder, custom GPT setups, and phone-based prompting.