Short answer: CodeJudge does not publish a classifier that reads your code and declares it AI-written. What it publishes is a seven-item violation list that a recruiter can filter your entire candidate pool by, and two of those items are unusual enough to matter: IP address change and geolocation change. Add multiple monitor detection, eye-movement image proctoring and a plagiarism percentage, and CodeJudge's model is a checklist of red flags sitting next to your name in the recruiter's queue.
Last checked: September 17, 2026. Every proctoring claim below links to CodeJudge's own documentation. We build InterviewMan, an AI interview assistant, so read the practical advice with that in mind.
Key takeaways:
- CodeJudge's documented violation filters are: "Left full-screen mode, Switched tab or window, IP address change, Geolocation change, Proctoring stopped, Multiple Monitor Detection and Plagiarism Percentage" (codejudge.io).
- Violations appear as "red flags next to their names", with a red "P" specifically for plagiarism.
- "Advanced AI-based Image Proctoring" watches eye movement and flags candidates who look away from the webcam, plus face verification and no-face or multiple-face detection.
- The Proctor report captures "tab-switching, screen-sharing, or opening new browser tabs", with periodic proctoring images viewable in a gallery or downloadable (codejudge.io).
- CodeJudge scores code quality across what it advertises as 127 or more code metrics, separately from correctness.
- There is no published AI-similarity engine, no lockdown browser and no process scan. Plagiarism matching is candidate-to-candidate.
What CodeJudge is
CodeJudge is a technical assessment platform used to screen, evaluate and interview candidates across IT and some non-IT roles. It supports what it describes as 20 or more coding languages and 50 or more technology frameworks, and it leans on project-based assessments rather than only short algorithm problems.
Its distinguishing pitch is measurement depth. CodeJudge advertises data-driven reports with insights across 127 or more code metrics, producing a code quality score alongside a correctness score. Its assessments product page lists plagiarism detection, AI image proctoring, code quality scoring, question pooling, custom assessments, video recording and code playback for interviews.
That code quality layer is worth flagging early, because it interacts with AI use in a way most candidates never consider. We come back to it below.
Ready to ace your next interview?
InterviewMan gives you real-time AI answers during live interviews — undetectable on Zoom, Meet, and Teams.
Try InterviewMan FreeWhat CodeJudge monitors
Leaving full-screen mode
CodeJudge assessments run in full-screen. Dropping out of it is a discrete, logged violation with its own filter. This catches more people than tab switching does, because exiting full-screen is what happens when you click anything else on your machine.
Switching tab or window
Logged separately from full-screen exit. The Proctor report "captures candidate behaviours during the assessment, such as tab-switching, screen-sharing, or opening new browser tabs", and the violation counter increments each time.
IP address change
This is unusual and worth understanding. CodeJudge flags a change of IP address mid-assessment.
Innocent causes exist: a home connection dropping to mobile tethering, a VPN reconnecting, an office network handing out a new lease. But the reason the flag exists is less innocent, and it is one of the strongest signals against remote assistance. If a session's IP changes, one plausible explanation is that a different machine or a different person picked it up.
Geolocation change
Same logic, different data source. CodeJudge flags a change in reported geolocation during the test.
Multiple Monitor Detection
CodeJudge detects and flags a second display. Given how many candidates keep documentation, notes or a helper open on a second screen, this is one of the higher-yield flags on the platform, and one of the easiest to trip innocently.
Proctoring stopped
If the proctoring layer stops running during your session, that itself is a flag. Killing the extension, denying camera permission mid-test or a browser crash all land in the same bucket, and the report does not distinguish sabotage from a flat battery.
AI-based image proctoring
CodeJudge describes "Advanced AI-based Image Proctoring" covering the candidate's eye movement, a flag when candidates look away from the webcam, face verification, and no-face and multiple-face detection.
Proctoring images are captured at intervals. In a recent update, "the candidate's webcam image will now be displayed separately on the right side of the screen", and images can be reviewed in gallery view or downloaded.
Plagiarism percentage
CodeJudge's plagiarism detector compares your submission against other candidates. In its own words, "When the code matches with other candidates then Plagiarised status is Yes", and the platform "Spots copied code, cheating during assessments and other malpractices". Its guidance on code quality and plagiarism covers how employers are meant to use it.
Note the framing: candidate-to-candidate matching. This is classic structural plagiarism detection, not AI authorship detection.
Code quality metrics
Separate from all of the above, CodeJudge scores the quality of your code across a large metric set: complexity, maintainability, structure and similar dimensions. This is not marketed as a cheating signal. In practice it functions as one, because generated code has a recognisable quality profile, and a submission that is stylistically immaculate but sits alongside a violation timeline tells a story.
What CodeJudge cannot see
| Signal | CodeJudge sees it | Notes |
|---|---|---|
| Leaving full-screen mode | Yes | Its own violation filter |
| Tab and window switching | Yes | Counted in the Proctor report |
| A change in your IP address | Yes | Filterable violation |
| A change in your geolocation | Yes | Filterable violation |
| A second monitor | Yes | Multiple Monitor Detection |
| Proctoring being disabled mid-test | Yes | Flagged as its own violation |
| Where your eyes are pointing | Yes, when image proctoring is on | Eye movement plus look-away flags |
| How many faces are in frame | Yes | No-face and multiple-face detection |
| Code matching another candidate | Yes | Plagiarism percentage |
| Code matching known AI output | Not documented | No published AI-similarity engine |
| Applications running on your desktop | Not documented | No process scan described |
| A phone out of camera frame | No | Camera inference only |
| Your network traffic contents | No | It sees the IP, not the payload |
Two caveats we repeat everywhere. Undocumented features exist, so "not documented" is not "impossible". And a reviewer scrubbing your timeline and your image gallery is not limited to what the software labelled.
How CodeJudge scores or flags suspicious behavior
CodeJudge does not produce a single integrity number like CodeSignal's Suspicion Score. It produces flags and filters.
The design detail that matters most to you is the filter. A recruiter with 300 submissions can filter the entire pool by any of the seven violation types, or by plagiarism percentage. That is a different risk shape from a score buried inside a report. It means your session does not need to be read to be excluded. It needs to match a filter.
Ready to ace your next interview?
InterviewMan gives you real-time AI answers during live interviews — undetectable on Zoom, Meet, and Teams.
Try InterviewMan FreeFlags surface visually too: red flags next to candidate names for rule violations such as multiple monitor use or IP address changes, and a red "P" for plagiarism.
So the realistic path to rejection on CodeJudge is not a classifier deciding your code looks generated. It is a recruiter filtering for "Multiple Monitor Detection = yes" and never opening your submission at all.
What recruiters see in the report
CodeJudge organises the candidate report into three tabs.
Timeline. "A chronological representation of a candidate's activity during an assessment", with "a detailed breakdown of the time taken by the candidate to complete various sections or questions within the test". Events can be filtered to track tab switches, full-screen entries and proctoring interruptions.
Proctor. Described as "a powerful tool that aids in maintaining test integrity", capturing tab-switching, screen-sharing and new browser tabs. It shows violation counts that increase with each infraction, the webcam image displayed separately on the right, and the periodic proctoring images in a gallery that can be downloaded.
Information. "A consolidated view of a candidate's profile details and their test performance": name, email, assessment duration, score, notes added during the process, and the resume when detail capture is enabled.
Alongside those sit the score, the code quality metrics and the plagiarism percentage. For interviews specifically, CodeJudge also offers video recording and code playback.
Using an AI interview assistant on CodeJudge: what works, what is risky
The rules first. If the employer's instructions prohibit outside assistance, using an AI assistant breaks them and a withdrawn offer is a real outcome. Read Should you use AI in a job interview in 2026? and check which companies allow or ban AI in interviews before you decide anything.
Clearly risky on CodeJudge, in order of how easily a recruiter can filter for it:
- A second monitor. Multiple Monitor Detection is a one-click filter. Whatever is on that second screen is irrelevant; the flag does not care.
- Leaving full-screen. Clicking any other window drops full-screen and produces a violation. This is the easiest flag to trip by accident and the easiest to avoid deliberately.
- Anything that stops the proctoring layer. Denying a permission, blocking a script or killing an extension mid-test produces "Proctoring stopped", which reads worse than the behaviour it was hiding.
- Connecting through a VPN that reconnects. IP address change is a flag with an innocent explanation that you will never get the chance to give.
- Pasting a solution. Plagiarism percentage compares against other candidates, and popular problems attract popular answers.
Where CodeJudge's documented signals do not reach:
CodeJudge's published proctoring is browser-level and camera-level. Nothing in it describes enumerating the applications running on your machine. An overlay drawn on top of the browser that never takes focus does not exit full-screen, does not register as a tab switch, and is not a second monitor.
That is the design reason our AI interview assistant is a native desktop application rather than a browser extension or a second window. On macOS and Windows the overlay is excluded from screen capture and recording APIs and hidden from the dock, taskbar and app switcher. Our interview copilot behaves the same way in live rounds. What each mode covers, and what it does not, is on our undetectability page.
Where we will be honest about the limits:
Image proctoring watches your eyes. Eye movement tracking and look-away flags are a camera problem, and no desktop software solves a camera problem. Our two-device Stealth Mode, which shows answers on a phone, is the poorest fit on a CodeJudge assessment with image proctoring on, and we would rather say that than sell you a mismatch. InterviewMan also does not type or submit answers for you, does not execute code and does not hide network traffic.
And there is a CodeJudge-specific point worth sitting with. Code quality scoring means the artefact itself is measured in unusual depth. Code you did not write and do not understand tends to be either suspiciously clean or subtly wrong in ways you cannot defend when the interviewer asks. Neither outcome is good.
For reference: our pricing is a free trial with no credit card, then $30 per month or $12 per month on the annual plan at $144 a year, with the stealth features on every plan. Apps for macOS, Windows, iOS, Android and the web are on the download page.
How to prepare for a CodeJudge assessment honestly
- Unplug the second monitor. Physically. Multiple Monitor Detection is the highest-yield filter on this platform and there is no argument to be made after the fact.
- Turn off the VPN. IP address change and geolocation change are both flags, and a VPN that re-handshakes mid-test produces them for no benefit.
- Stay in full-screen. Close everything before you start so nothing pulls you out of it: notifications, calendar reminders, update prompts.
- Grant camera permission before the timer starts. A permission prompt handled badly mid-test can produce "Proctoring stopped".
- Write clean, readable code deliberately. Code quality is scored on its own, so naming, structure and simple functions earn real points here that they do not earn on a pure test-case platform.
- Expect to explain it. CodeJudge supports video interviews with code playback, so the submission is often the start of the conversation rather than the end.
- Do the underlying work. Our interview prep hub, technical interview preparation guide and how to pass a coding interview in 10 steps cover the parts that move your score.
How CodeJudge compares to CodeSignal, HackerRank, CoderPad, and Codility
| Platform | Named AI detection | Webcam proctoring | Second monitor detection | IP or location flags | Code quality scoring |
|---|---|---|---|---|---|
| CodeJudge | Not documented | Yes, AI image proctoring with eye movement | Yes, a filterable flag | Yes, IP and geolocation change | Yes, 127 or more metrics |
| CodeSignal | Yes, inside Suspicion Score | Yes, on proctored assessments | Not published as a filter | Telemetry informs the score | Not a headline feature |
| HackerRank | Yes, ML model plus MOSS structural matching | Optional | Not published as a filter | Not published as a filter | Not a headline feature |
| CoderPad | No named classifier | Enterprise plans only | No | No | No |
| Codility | AI detection layered onto Similarity Check | Optional | Not published as a filter | Not published as a filter | Partial, via task scoring |
Sources: HackerRank on plagiarism detection, CodeSignal on Suspicion Score, Codility on plagiarism prevention and fraud detection, CoderPad's cheating prevention documentation.
CodeJudge's genuine advantage is that it tells candidates and employers exactly what it measures. Seven named flags, a plagiarism percentage, a published metric count. There is no black box deciding your code looks generated, which means there is no black box mislabelling a strong candidate either. The trade-off is that filterable flags are unforgiving: an innocent VPN reconnect looks identical to a suspicious one.
CodeJudge's own overview of remote proctoring is a fair read on how the vendor frames it, and our broader survey is in how companies detect AI cheating in interviews.
Frequently asked questions
Does CodeJudge detect ChatGPT?
Not through a published AI-authorship classifier. CodeJudge's plagiarism detection compares your code against other candidates, not against known model output. What it does instead is flag the behaviour around AI use: leaving full-screen, switching tabs, a second monitor, and eye movement away from the webcam.
What violations does CodeJudge flag?
Seven, and they are all filterable by the recruiter: left full-screen mode, switched tab or window, IP address change, geolocation change, proctoring stopped, multiple monitor detection, and plagiarism percentage. Violations show as red flags next to your name, with a red "P" for plagiarism specifically.
Does CodeJudge detect a second monitor?
Yes. Multiple Monitor Detection is one of the named violation filters. It flags the presence of a second display regardless of what is on it, so disconnecting an unused external monitor before you start is worth doing.
Why does CodeJudge track my IP address?
An IP change mid-assessment can indicate that a different machine or a different person took over the session. Innocent causes exist, such as a VPN reconnecting or a home connection failing over to mobile data, but the flag does not distinguish between them, so avoid a VPN during the test.
Does CodeJudge use a webcam?
When image proctoring is enabled, yes. CodeJudge describes advanced AI-based image proctoring covering eye movement, a flag when candidates look away from the webcam, face verification, and no-face and multiple-face detection. Proctoring images are captured at intervals and shown to recruiters in a gallery.
What is CodeJudge's code quality score?
A separate score from correctness, computed across what CodeJudge advertises as 127 or more code metrics covering things like structure, complexity and maintainability. It is not marketed as a cheating signal, but it does mean the readability and shape of your code affects your result, not just whether the tests pass.
Can CodeJudge see what applications I have open?
Nothing in its published documentation describes scanning the processes or applications running on your machine. Its documented signals are browser-level, network-level and camera-level. That is not a guarantee about undocumented behaviour, and image proctoring can still catch you looking somewhere else.
What happens if my proctoring stops mid-test?
It becomes its own violation, recorded as "Proctoring stopped". The report does not distinguish a deliberate interference from a browser crash or a denied permission, so the safest approach is to grant every permission before the timer starts and avoid touching browser settings during the assessment.
See also
Ready to Ace Your Next Interview?
Join 570,000+ professionals using InterviewMan to get real-time AI assistance during their interviews.
