How AgentR Stops Cluely and AI Interview Cheating, Fairly

Interview cheating has moved from proxy candidates to hidden AI tools like Cluely, which interviewers can’t see on a shared screen. AgentR handles it in two ways: privacy-first proctoring that detects these tools at system level, and an AI interviewer whose follow-ups build on each candidate’s own answers, which makes outside help slow and obvious. Honest candidates are never scored on how they look or speak.

Tools like Cluely, Parakeet AI, Interview Coder, Final Round AI and LockedIn AI listen to the interview and show answers in a window that doesn’t appear when the candidate shares their screen. We’ve written about how these overlays work.

The scale is real. In a survey of 3,290 job candidates, Gartner found 39% had used AI during the application process. For Indian IT teams, this lands on top of a familiar problem: proxy interviews and fake candidates were already costing engineers hours of interview time. Some large employers, including Google, have responded by bringing interviews back in person, which is slow and expensive at volume.

What happens when a candidate opens Cluely in an AgentR interview?

AgentR Guard spots it and pauses the interview until the tool is closed. Guard is a small desktop app for macOS and Windows that candidates install before a proctored interview, whether it’s an AI first round or a round your own team runs on any video call software.

1. The candidate opens the interview link and starts AgentR Guard. It runs only for the length of the session.
2. Guard checks what is running on the computer throughout the interview.
3. If Cluely or a similar tool opens, the interview pauses until it is closed.
4. Once the computer is clean, the interview picks up where it left off.
5. The recruiter gets one read on the session, from “clean” to “worth a look”, and decides what it means.

Why a desktop app instead of a browser check?

Many proctoring tools avoid an install and run in the browser instead, often with webcam gaze tracking. That is easier to roll out, but a browser tab can’t see other apps on the computer, and a webcam can only guess from where someone looks. A 30fps webcam can tell roughly whether a candidate is facing the screen, not what they are reading or whether they are just thinking.

Sees tools hidden from screen share (Cluely, Parakeet)

• In-browser proctoring: No. A browser can’t see other apps.
• Webcam gaze and behaviour scoring: No. It guesses from eye movement.
• AgentR Guard (desktop app): Yes. It checks what is running on the computer.

Sees a second monitor

• In-browser proctoring: Partly, with permission.
• Webcam gaze and behaviour scoring: It guesses from glances.
• AgentR Guard (desktop app): Yes.

Risk to honest candidates

• In-browser proctoring: Low.
• Webcam gaze and behaviour scoring: It can misread thinking, nerves, glasses, poor lighting or disability.
• AgentR Guard (desktop app): Low. It never looks at the candidate.

Sees a phone or second device

• In-browser proctoring: No.
• Webcam gaze and behaviour scoring: It guesses from glances.
• AgentR Guard (desktop app): No. The AI interview covers this.

Candidate setup

• In-browser proctoring: Nothing to install.
• Webcam gaze and behaviour scoring: A webcam and good lighting.
• AgentR Guard (desktop app): One install, macOS or Windows.

What the recruiter gets

• In-browser proctoring: Tab-switch logs.
• Webcam gaze and behaviour scoring: A behaviour score.
• AgentR Guard (desktop app): One read, from “clean” to “worth a look”.

No single check catches everything. Guard covers what a browser can’t see, and the AI interview itself covers what no software on the laptop can, like a phone.

How does the AI interview catch what software can’t?

The best defence against cheating is an interview where it doesn’t pay off. AgentR interviews ask candidates to do the work instead of describing it: draw the system, handle the pushback, fix the real problem in front of them. The AI interviewer also listens to each answer and builds its next question on what the candidate actually said, which a script prepared in advance can’t anticipate.

We incorporate several other quiet safeguards into our interview process. The effect is clear; candidates relying on external screens tend to fall apart on the second or third follow-up. Ultimately, securing the environment and asking the right questions together raise the cost of cheating far more than either does alone.

Our goal: protect honest candidates, make it hard for imposters

A cheating check that catches a cheater but also flags a nervous fresher has failed. In a market where thousands of honest candidates go through the same first round, that cost is real. So AgentR follows five rules:

• We look at tools, never at faces. No scoring of eye movement, facial expressions or speaking speed. Looking away while thinking is normal, and an accent is not a red flag. Detecting the tool itself is fairer than guessing from behaviour.
• One signal never fails anyone. A single odd moment gets logged and nothing more. Only several signals that back each other up raise a concern.
• A person decides. Guard pauses and marks. It never rejects a candidate.
• Candidates know it’s there. That matters for consent under India’s DPDP Act, too.
• Guard only runs during the interview. It starts with the session, stops when it ends, and recruiters see one read on the session, never a log of what is on the candidate’s computer.

What doesn’t AgentR catch?

Guard sees what runs on the candidate’s computer, so anything off the computer is outside its reach. That is why the AI interview matters as much as the proctoring.

• A phone or second device. Guard can’t see it. Follow-ups built on the candidate’s own answers make relayed help slow and obvious instead.
• Someone else in the room. Guard checks software, not the room around the laptop.
• Candidates without a Mac or Windows computer. Guard runs on macOS and Windows only, so plan another route for them, such as an in-person round.

What should you ask any cheating-detection vendor?

Ask four things: how the score is worked out, what happens to honest candidates, who reviews a flag, and what candidates are told. Most cheating-detection pages buyers are pointed to skip at least one of them:

• How the score was worked out. A “cheating probability” or accuracy percentage with no method behind it tells you very little.
• What happens to honest people? Many tools lean on gaze tracking, response timing, or speech patterns, which can misread nervous candidates, non-native speakers, and people with disabilities. Wrong flags are rarely discussed.
• Who reviews a flag. “Human review” gets a passing mention, without saying who looks or whether a flag alone can reject someone.
• What candidates are told. Consent, notice, and data retention, including DPDP, are mostly absent.

Ask every vendor about all four, AgentR included. Better still, ask them to show it working on a laptop you choose.

What does it cost?

Nothing extra. As of October 2026, proctoring and verification are included on every AgentR account. The plan is ₹11,999 a month, which becomes ₹12,000 of usage at ₹25 per applicant reviewed and ₹10 per interview minute (prices exclude GST). Your first role is free: up to 50 applicants and 5 proctored interviews, no card needed.

Questions people ask

Can Cluely, Parakeet, or similar interview tools be detected in an interview?

Yes, with the right setup. Cluely, Parakeet, Interview Coder and similar tools are built to stay out of screen shares, so watching a shared screen on Zoom, Meet, or Teams won’t show it. It takes software on the candidate’s computer that monitors at system level. AgentR Guard does that and pauses the interview when it detects Cluely or a similar tool.

Will an honest candidate get flagged for looking away or pausing?

No. AgentR doesn’t score eye movement, pauses, or speaking speed, and no single signal can flag or fail a candidate. Anything that is marked goes to a person.

Do candidates have to install anything?

Yes, for proctored interviews. AgentR Guard runs on macOS and Windows, and only while the interview is running.

Why not use a browser-based check instead?

A browser tab can’t see other apps on the computer, so it misses overlay tools like Cluely by design. Webcam gaze tracking can only guess from where a candidate looks. The comparison above shows what each approach can and can’t see.

Does it work for interviews our own team runs?

Yes. Guard can run alongside a human-led interview on any video call software. It checks the candidate’s computer, and your interviewer runs the conversation as usual.

Try it on your next open role. The first one is on us, proctored interviews included.

See what’s included on the AgentR pricing page, or read more about how AgentR prevents AI cheating.