What CoderPad really records when a candidate uses AI

The short answer: CoderPad flags behaviour, not authorship

It helps to separate the two CoderPad products, because they log different things. CoderPad Interview is the live pad, with an interviewer watching. CoderPad Screen is the asynchronous test you take alone before a deadline. If you have not used either yet, our guide on what to expect in a CoderPad interview covers the format. This post is only about detection.

• Both products keep a playback of how the code was written, mark pastes from outside the editor, log time spent away from it, and save everything typed into the built-in AI Assist panel.
• Interview only: live alerts to the interviewer, a copy-all alert, optional video and audio in the pad and, on some plans, an AI notetaker that transcribes the call.
• Screen only: exact-match plagiarism checks, flags for abnormal speed, IP-based location tracking, optional full-screen mode, optional paste blocking, an optional AI-generated follow-up question and, on Enterprise plans, webcam proctoring.

What is missing from that list matters just as much. CoderPad documents no classifier for AI-written code, no scan of the other apps on your computer and no way of seeing a second device. Both cheating-prevention pages also repeat one caveat: a flag does not mean cheating, and someone still has to look at it.

CoderPad playback: the keystroke record

Playback is the main detection tool, and it is passive. According to CoderPad's playback documentation, the Interview platform records every keystroke of every participant. When the interviewer clicks End Interview, the pad switches to Playback Mode. Each participant gets their own track showing edits (a few keystrokes each) and code runs. The reviewer can scrub back and forth, replay at 1x to 8x, and watch playback switch between tabs and files in the order they happened.

One limit is documented. Playback shows only the code changes in the editor, and the docs say that anything run in the shell, such as REPL commands or package installs, does not appear.

CoderPad's cheating-prevention guide for Interview tells interviewers what to look for when they replay a session: "bursts of activity inconsistent with normal typing", perfect code arriving after a long silence, and long pauses before follow-up answers. This is the closest CoderPad comes to detecting AI. It does not identify the source. It shows that 40 finished lines appeared after two quiet minutes, and a person has to interpret that.

In Screen, code playback covers programming, game and free-text questions and colours the timeline by activity: green for running or submitting code, orange for writing, red for pasting, dark grey for leaving or returning to the test and light grey for time spent outside it. Reviewers can watch at 0.5x to 8x and rerun your code themselves.

Paste detection and copy-all alerts

Pasting is the easiest signal to catch, so CoderPad logs it in several ways.

• Interview playback. Each paste from outside the editor puts a yellow dot on the playback bar. Hovering over it shows "Code pasted", and clicking jumps to that moment.
• Live alerts. The Interview guide says CoderPad alerts the interviewer during the session when a candidate pastes code from an external source.
• Copy-all. The playback docs describe an alert that fires "when a candidate copies all the code in a file". It repeats for each file or each time it happens, and it stays in the Pad Summary.
• Screen. Pastes show as red segments, and the reviewer can see what was pasted. The question statement can never be copied, and the company can block pasting from outside the IDE completely. CoderPad itself notes that this has "a lot of legitimate use cases".

One detail surprises people. The playback docs say that copying from CoderPad's own AI Assist tab can still trigger the paste notification. A pasted block is flagged whatever its source, which is why CoderPad keeps reminding reviewers that a paste is not proof of anything.

CoderPad tab switching: what "left the pad" means

In Interview, leaving the editor shows up as an orange dot on the playback bar, and the tooltip gives the duration. CoderPad's own example reads "Clicked away for 19 seconds". The interviewer also gets a live alert when a candidate leaves the IDE, and once the pad ends the Interview Summary tab lists left-pad events and external pastes under suspicious behaviour, with a total for each. Opening the browser console also counts as leaving, because the docs say the console is technically a different browser tab.

Notice what is and is not recorded. The documented signal is that the editor lost browser focus, plus how long for. CoderPad's pages do not say it records where you went or what you looked at, and they do not describe watching other applications. A two-second click on a notification and a two-minute visit to a chatbot both show up as an absence with a timestamp. The reviewer reads the pattern and the duration, then often asks you about it.

Screen can be stricter. The Screen cheating-prevention page describes an optional full-screen mode in which leaving full screen or switching monitors triggers an alert after a 10-second grace period. Time outside the test appears as grey segments in playback.

CoderPad Screen: plagiarism, speed, location and webcam

Screen tests have no live observer, so CoderPad adds checks after the fact. The Screen documentation lists them:

• Plagiarism. Submissions that exactly match another candidate's code are flagged. CoderPad's anti-cheat product page describes it as detection "when a previous candidate's exact code is reused" and says CoderPad cannot publicly share how the engine works. Details are offered to customers under an NDA. So there is no public evidence that it compares your code with AI output in general.
• Abnormal performance. Difficult questions solved in "a fraction of the usual time" are flagged.
• Geolocation. Approximate location is tracked by IP address to catch logins from different places or devices. The docs note that a VPN or travel can explain a change.
• Question protection. Random question blocks mean tests differ between candidates, and CoderPad says it watches the web for leaked questions and replaces them.
• AI follow-up question. CoderPad's test settings describe a beta feature in which ChatGPT writes a follow-up question asking you to explain part of your own code, then checks the answer. Questions are more likely on exercises where suspicious activity was detected, and the docs advise reviewers to check the validation because it "can make mistakes from time-to-time".
• Webcam proctoring. CoderPad's webcam proctoring docs say this is available only on Enterprise plans on request. Periodic snapshots can be analysed by AI for three patterns: no face, multiple faces or a different face. You can take the test with the camera off, but the report will say so, and snapshots are kept for at most 90 days.

On the reviewer's side, flagged candidates get a red exclamation mark in the candidate list. The simplified report shows plagiarism, IDE-exit and geolocation alerts, and the detailed report breaks issues down by question. A recruiter who marks "Candidate cheated on this exercise" removes the points for that question and adds a Cheated label.

CoderPad AI detection inside the pad: AI Assist is logged

CoderPad's answer to AI is mostly to bring it inside the pad, where it can be seen. The AI Assist documentation describes a chat panel connected to models such as GPT, Claude, Gemini and Llama. An organisation setting decides whether it starts on or off, and the interviewer can switch it for each pad. In shared mode both of you see the prompt history. There is also a private mode the candidate does not see, which the interviewer can use to analyse your code. The line that matters for detection is this one: "All pad users' prompts and AI output will be saved for review in playback mode."

In Screen, AI Assist is available only for Project questions, when the company enables it, and the prompts and output are saved in playback in the detailed test report. The Interview guide tells interviewers to enable AI Assist "so all AI usage is visible within the platform", and to ask candidates to explain and critique what the model produced.

What employers allow varies. Our comparison of AI-assisted interviews on HackerRank, CodeSignal and CoderPad covers the official policies. For Screen specifically, CoderPad's candidate preparation guide says that for programming questions you can use "all available resources like StackOverflow or Google", but you "may not have someone else actively solve the question for you". Outside AI is not mentioned. The company's own instructions are what count.

What CoderPad cannot see, and what the interviewer still can

Based on the pages cited here, these things are not documented as detected:

• Which other application or website you used. The logs record that you left and for how long, not where you went.
• A phone or second computer next to you. Nothing in the docs refers to other devices, apart from the IP-based location check in Screen.
• Whether finished code came from a model, a memory or an old project. There is no documented AI-text classifier.
• Commands typed in the shell, which the docs say are not shown in Interview playback.

None of this means a live round is unobserved. The interviewer is the detection layer CoderPad leans on most. The Interview guide suggests turning on video and audio "to confirm the candidate is the one coding". CoderPad's video does not allow backgrounds or background filters. For higher-security interviews the guide says the interviewer "may also ask the candidate to share their screen if off-pad activity is suspected". On Enterprise and Custom plans, the AI Notetaker can transcribe the call and produce AI notes, but each participant must opt in before their voice is transcribed. The same guide names the human signal that matters most: a mismatch between how fluently a candidate talks and how deep their technical answers go. CoderPad's advice is to "treat alerts as prompts for clarification, not immediate disqualification", which in practice means the interviewer asks you to explain what you did.

Whether AI is allowed at all is the employer's decision. See which companies allow AI in interviews, with sources.

Where a screen-capture-excluded overlay fits, and where it does not

Some desktop tools, including our own AI coding interview assistant, draw their window in a way the operating system leaves out of screenshots and screen shares. On Windows this is a documented API flag: Microsoft's SetWindowDisplayAffinity reference says a window set to WDA_EXCLUDEFROMCAPTURE "does not appear at all" in captures. Microsoft also says plainly that this is not a security feature and gives no guarantee, for example against someone photographing the screen.

Here is what that changes and what it does not. It affects whether a window appears when you share your screen or someone records it. It does nothing to what CoderPad logs inside the pad. Keystroke timing still goes into playback. A paste is still a paste. A click outside the editor is still a focus change with a duration. Plagiarism, speed and location checks in Screen run on your submission, not on your desktop. A follow-up question, whether from ChatGPT in Screen or from the interviewer in a live pad, still needs you to explain the code yourself. Interview Copilot listens to the interview audio and suggests answers, and it can work through a coding problem from a screenshot. None of that is hidden from CoderPad's own records, and we do not claim it is undetectable.

One more honest point. Detection is only part of how you are assessed. Interviewers decide mostly on how you reason: why you chose this approach, what it costs, what you would test next. A solution you cannot explain fails that test whatever any log says. The rules are set by the employer, not by CoderPad or by us. Some companies switch on AI Assist and want to watch you use it, while others forbid outside help completely. If you are unsure, ask at the start of the session. Following the rules you were given is your responsibility.

Taking another platform? See also: can HireVue detect AI, can CodeSignal detect AI, can Codility detect AI, can HackerRank detect AI.

Sources

• CoderPad Docs: Cheating prevention in Interview
• CoderPad Docs: Playback mode
• CoderPad Docs: Cheating prevention and detection in Screen
• CoderPad Docs: Code playback in Screen candidate reports
• CoderPad Docs: Editing test settings (AI follow-up question, paste blocking)
• CoderPad Docs: Webcam proctoring
• CoderPad Docs: AI Tools for Interview
• CoderPad Docs: AI Tools for Screen
• CoderPad: Cheating prevention product page and FAQ
• CoderPad Docs: Live interview transcription
• CoderPad Docs: Screen preparation guide for candidates
• Microsoft Learn: SetWindowDisplayAffinity

Every factual claim above was checked against these pages on 31 October 2026. CoderPad changes its features and plan tiers, so check again before your test.

Related articles