Is Cluely Detectable in 2026? What Detection Tools Actually See

Short answer: it depends entirely on who is watching and how. On a plain Zoom, Meet, or Teams call, a correctly built native overlay is genuinely invisible to the screen share — the screen share only carries pixels, and a properly excluded window never becomes a pixel the interviewer receives. But in a proctored coding assessment, the picture flips completely. Several assessment vendors now say, in their own documentation, that they detect and block Cluely — and they do it without ever looking at your screen share.

That distinction is the whole story, and almost no page ranking for this question makes it. So let’s be precise about it, with sources you can check yourself.

Here is the reframe that matters: working detection does not hunt for the overlay in your screen share. It reads the list of programs running on your machine, or it watches how you behave. Once you understand that, the marketing claim of “100% undetectable” falls apart — not because the invisibility is fake, but because it is answering the wrong question.

The two threat models (don’t let anyone blur them)

	Plain video call (Zoom / Meet / Teams)	Proctored assessment (locked-down desktop app)
What it captures	Screen pixels + your camera/mic	Your process list, plus it can block apps from launching
Can it see a correctly-excluded native window?	No — the compositor drops the window before pixels reach the call	Not via pixels either, but it doesn’t need to — it reads running processes
Named examples that target Cluely	(none — a call app doesn’t inspect processes)	HackerRank Desktop App, Proctorio, Honorlock
Your real exposure	Behavioral tells (eyes, timing, cadence)	The tool being enumerated or blocked on your machine

If you take one thing from this page: a Zoom interviewer sees pixels, not processes. A proctoring app sees processes, not just pixels. Those are different worlds, and “is Cluely detectable” has a different answer in each.

Who is actually hunting Cluely — and how

The interesting part is that the people trying to catch tools like Cluely mostly gave up on the screen share. Here is what they publish about their own methods.

Proctoring vendors that name Cluely in their own docs

These are the strongest receipts, because they are the vendors’ own words:

• HackerRank. Its Desktop App Mode knowledge base states verbatim that the app “detects and closes any other programs that candidates try to open during the test, including invisible cheating tools such as Cluely and Ultracode” (HackerRank support KB). Note the mechanism: it closes programs — it is reading your process list, not scanning pixels.
• Proctorio. Its blog explains that it “blocks tools like Cluely by preventing unauthorized apps from launching during exams,” using a Secure Companion App. Proctorio openly acknowledges that Cluely “uses system-level shortcuts and overlays that do not appear to traditional screen-sharing tools” — i.e. their approach is app-launch prevention, not pixel detection (Proctorio blog, May 2025).
• Honorlock. Its companion application can “block Cluely (and all other applications) during exams.” Honorlock itself describes Cluely as “almost undetectable” via screen share — an admission that pixel-level detection is the wrong tool for the job (Honorlock blog, April 2026).

The honest caveat here — and it is a big one — is that all three of these are exam/coding-assessment proctoring tools. They are a locked-down desktop client the company makes you install and run. On a normal Zoom or Teams interview, there is no such companion app inspecting your machine. Don’t let anyone tell you “Proctorio can block Cluely” means “your Zoom interviewer can see Cluely.” Those are different setups.

The dedicated anti-Cluely tools

When Cluely blew up, a small detection industry appeared around it. In April 2025, TechCrunch reported that Validia launched “Truely,” a free tool that “triggers an alarm if it detects someone using Cluely,” and that a company called Proctaroo claims to catch it the same way. Proctaroo’s CEO Adrian Aamodt is quoted directly:

“When a Proctaroo session is active, we can see running applications and ‘hidden’ background processes — Cluely is no different.”

Again: running applications and hidden processes. Not the screen share. These tools work by asking you to run a separate app during the interview that inspects your own machine — something a screen share would never reveal.

Cluely’s own co-founder, Roy Lee, dismissed all of this and signaled that the future is hardware, telling TechCrunch: “Whether it’s smart glasses, a transparent glass screen overlay, a recording necklace, or even a brain chip, we’re not sure.” That is, in effect, an acknowledgment that software overlays are a cat-and-mouse game against process inspection.

It’s worth remembering how this category started, because it is documented on the record. Cluely is the rebrand of Interview Coder, built by Columbia student Chungin “Roy” Lee. He posted a video of himself using it to pass an Amazon technical interview; Amazon filed a Behavioral Conduct Reporting Form with his university, and Columbia suspended him (Gizmodo; Columbia Spectator). The category’s origin story is, literally, someone getting caught — because he published the evidence himself.

What a plain video call can — and cannot — see

Here is the part that the “100% undetectable” marketing gets right, technically, and the part it quietly overstates.

A screen share is a pixel capture of a display or a window. The video-call app asks the operating system for the composited pixels of whatever you are sharing (via ScreenCaptureKit on macOS, or the Desktop Duplication / Windows Graphics Capture APIs on Windows) and sends those pixels down the call (Adam Svoboda’s technical writeup, August 2025).

A native window can ask the OS to remove itself from that capture. On macOS a window sets NSWindow.sharingType = .none; on Windows it calls SetWindowDisplayAffinity with the WDA_EXCLUDEFROMCAPTURE flag (Microsoft docs). When set, the compositor drops the window before the pixels are handed to Zoom or Teams — so the call literally cannot recover it through the screen share, regardless of settings.

And crucially: a video-call app does not enumerate your running processes. Zoom, Meet, and Teams capture pixels; they do not walk your process list to see what apps you have open. That is precisely why the people trying to catch you migrated off the screen share and onto (a) companion proctoring apps that read your processes, and (b) watching the video itself.

But native exclusion is not magic, and honesty requires the caveats:

• It fails on some capture paths. Svoboda found first-hand that “QuickTime’s screen recorder bypassed the protection entirely,” and he notes “DirectX, kernel-level captures, or GPU frame grabs might still see you.”
• Microsoft’s own documentation warns that SetWindowDisplayAffinity is not a security or DRM guarantee — a photograph of your screen still captures everything on it.
• Exclusion only applies to native windows. A browser-extension overlay is drawn inside the page’s own pixels, so it has no separate OS window to exclude — which is why extension- and browser-tab-based copilots leak in screen shares while properly built native windows don’t. (More on that in Is LockedIn AI Detectable?.)

So on a plain video call, a correctly-excluded native overlay is genuinely invisible in the screen share. That is real. What is not real is the leap to “therefore you cannot be caught.”

The tells that have nothing to do with your screen

Even a perfectly invisible overlay doesn’t change how you behave while you read it. This is where the honest risk lives, and two independent sources describe the same tells.

The detection vendor Fabric — an AI-interview platform that sells cheating detection — published a technical breakdown of the behavioral signals it looks for. Because these numbers come from Fabric’s own platform and Fabric has a commercial incentive, treat them as vendor claims, not established facts. Fabric claims a study of 19,368 interviews and an 85% detection rate (Fabric study) — again, its own number, not independent research. The useful part is the description of the tells (Fabric deep-dive):

• Response timing. Fabric describes “a consistent 3 to 5 second delay after every question, regardless of difficulty” — abnormally low variance across easy and hard questions.
• Eye movement. “Reading eyes move differently than thinking eyes” — reading tracks in horizontal lines with rapid return saccades, versus the scattered movement of genuine thought.
• Speech. As the interview goes on, “vocabulary becomes more uniform, sentence structure regularizes, filler words disappear” — a drift toward suspiciously polished delivery.

You don’t have to trust a detection vendor to find this credible, because an unrelated first-hand human account describes exactly the same signals. In a moderated advice column, an interviewer wrote about catching a candidate: the giveaways were “the eyes reading and the delay before responding,” answers that restated the question and were “full of buzzwords with no substance,” and a cadence “like reading from a script” (Ask a Manager, November 2025). Two unrelated sources describing the same tells is about as solid as this space gets.

The lesson: the invisibility of the window was never the weak link. The reading cadence, the flat answer latency, and the generic scripted phrasing are.

One more thing about Cluely specifically

Two facts about the product itself are worth knowing:

Stealth is a $149.99/month upsell. As of July 2026, Cluely’s pricing page lists a Free tier, Pro at $19.99/month, and “Pro + Undetectability” at $149.99/month — where the top tier’s differentiator is being “Completely hidden to meeting screen sharing software” (Cluely pricing, accessed July 2026). The invisibility you’re reading about is a paid add-on, roughly 7.5x the base price. Cluely has also repositioned as an “Undetectable AI for Meetings” rather than an interview tool (cluely.com, accessed July 2026).

The client has had security scrutiny. Security researcher Jack Cable reverse-engineered Cluely and reported finding the app running with no sandbox, complete system prompts shipped in plaintext inside the client, plus a postMessage/IPC handler flaw he said could let a website opened through Cluely capture screenshots without the user’s knowledge. Cluely initially sent a DMCA takedown over his post; the employee who filed it later said they had done so without leadership sign-off, and the CEO apologized (Jack Cable on X; Critical Thinking podcast Ep. 136, corroborated independently by Hacktron AI). Whether that matters to you is a personal call, but “the tool is invisible” and “the tool is trustworthy with your data” are separate questions.

Your interview data lives in its cloud. Cluely’s privacy policy says it collects audio, transcriptions, and screenshots. In fairness, it also states that it does not sell or train on your data — a real point in its favor. But end-to-end encryption is opt-in (you email support to turn it on) rather than the default, and your content still passes through third-party transcription and advertising providers. Invisible on screen is not the same as private off screen.

So — is Cluely detectable?

Honestly:

• In a plain video interview (Zoom, Meet, Teams), a correctly-excluded native overlay is not visible in the screen share. The realistic risk is behavioral — your eyes, your timing, your cadence — not the pixels.
• In a proctored assessment run through a locked-down desktop client (HackerRank Desktop App, Proctorio, Honorlock, or a dedicated tool like Truely/Proctaroo), the overlay can be detected or blocked outright — because those tools read your process list, which no screen share does.

The problem with “100% undetectable” is not that the invisibility is fake. It’s the absolutism. Any claim that survives every capture path, every proctoring companion, and every behavioral analysis at once is overselling. The tools that make that claim are answering the screen-share question and hoping you don’t ask the other two.

Where Dusky stands

We’ll be equally plain about our own product, because pretending otherwise would make everything above worthless.

Dusky is a native desktop app that registers as a screen-share-excluded window at the OS level — the macOS system content protection API and the Windows display affinity API — so video-conferencing apps skip the window when you share your screen (stealth mode docs). It’s the same class of native OS exclusion described above, not a browser extension or a Zoom plugin, so there’s no in-the-page overlay to leak. Invisibility is always on — there is no stealth toggle to leave in the wrong position before a call.

What we will not tell you:

• We do not claim Dusky is invisible to proctoring companion apps that inspect your process list, to physical or external cameras, or to eye-tracking. The site is deliberate about this.
• We tell every user to test invisibility before every real interview — Dusky ships a built-in invisibility test during onboarding, and we recommend re-running it. Exclusion can be affected by old OS builds, certain capture tools, and corporate MDM/group-policy restrictions.
• On Windows, our build currently ships unsigned, so SmartScreen may warn you on install (choose More info → Run anyway). The macOS build is signed and notarized.

On data, our posture is pass-through: Dusky does not store your transcripts, screenshots, or conversations on our servers, and your Interview Context (CV and job description) stays on your machine — stored locally, AES-256-GCM encrypted, and wiped when you sign out. Invisibility and privacy are different promises, and we try to be plain about both.

If you want to see it rather than take our word for it, here’s the screen-share test:

And nothing above changes the behavioral reality: a copilot helps you structure what you already know — it doesn’t fix reading-off-a-script cadence for you.

Try it yourself — free, no credit card. Download Dusky and run its built-in invisibility test on your own machine before you ever rely on it. The trial gives you 15 minutes of AI-assistance time — the clock only counts while Dusky is actively answering, not while it sits idle — with every feature unlocked. Get it at getdusky.app.

For more, see Dusky vs Cluely, the honest Cluely alternatives rundown, and our own Is Dusky Detectable? page where we apply this same scrutiny to ourselves.

FAQ

Can a Zoom or Teams interviewer see Cluely on my screen?

Not through the screen share, if the overlay is a correctly-excluded native window — the operating system drops that window before the pixels ever reach Zoom or Teams (technical explainer). The realistic risk on a plain video call is behavioral: reading cadence, response timing, and scripted-sounding answers (Ask a Manager).

How do proctoring tools detect Cluely if it’s invisible on screen?

They don’t use the screen share. HackerRank’s Desktop App “detects and closes… invisible cheating tools such as Cluely and Ultracode” (HackerRank KB); Proctorio and Honorlock block it with a companion app that prevents unauthorized programs from launching. All of these read your process list or control what can run — something a normal video call never does.

Is any AI interview tool “100% undetectable”?

No tool can honestly promise that. Native screen-share exclusion is real and verifiable, but it doesn’t cover proctoring apps that inspect your processes, phone photos of your screen, certain capture paths, or behavioral tells. Absolute “100%” claims — common across this category — overstate what any overlay can do.

Does Cluely cost extra to be undetectable?

As of July 2026, yes. Cluely’s pricing page lists undetectability on its “Pro + Undetectability” tier at $149.99/month, versus $19.99/month for base Pro (Cluely pricing). Prices change, so re-check the live page.

What actually gets people caught in interviews?

Based on a detection vendor’s described signals and a first-hand interviewer account, the recurring tells are the same: eyes visibly reading, a consistent multi-second pause before every answer, and generic, buzzword-heavy, scripted-sounding responses (Fabric, attributed as a vendor claim; Ask a Manager). Note that the Fabric statistics are the vendor’s own numbers, not independent research.

How is Dusky different from Cluely on this?

Dusky uses the same class of native OS-level screen-share exclusion, but it’s a single native app with no browser extension or in-page overlay to leak, and invisibility is always on with no toggle to forget. We also don’t claim invisibility to proctoring apps, cameras, or eye-tracking, and we ask you to test before every interview. See Dusky vs Cluely and Is Dusky Detectable?.