What HackerRank can actually see when you use AI on a test
Can HackerRank detect AI? On 31 October 2026 we read HackerRank’s own help center and its 2026 release notes to answer that properly, instead of repeating what tool vendors claim on their landing pages. The honest answer depends on which of HackerRank’s integrity modes the employer switched on, and there are four of them. Some of them watch your browser tab. One of them watches your whole computer.
A lot of what ranks for this question is written by companies selling an interview tool (or a detection tool), and it tends to lean whichever way pays. We make Craqly, a desktop interview assistant, so read this with that in mind. We’ve tried to quote HackerRank directly wherever we can and to flag the places where its documentation simply doesn’t say.
Can HackerRank detect AI-generated code at all?
Yes, in two different ways. HackerRank’s code analysis scores behavior rather than naming a tool: how your code evolved, what you pasted, how long you left the tab, and how similar your answer is to others. Separately, in Proctor Mode, its screenshot analysis now looks for suspicious applications on screen, which HackerRank says includes newer AI tools. Flags go to a human reviewer. They’re signals, not verdicts.
The clearest public statement is HackerRank’s own June 2023 launch post for AI-powered plagiarism detection. It says the model tracks “dozens of signals” and claims a 93% accuracy rate. It also says the system “repeatedly detects ChatGPT-generated solutions, even when those results are typed in manually.” That last line matters more than the percentage, because retyping AI output by hand is the first thing most people try.
The same post says hiring managers can replay a candidate’s answer keystroke by keystroke. Think about what that looks like from the reviewer’s chair. Real problem-solving is messy: a half-written loop, a deleted helper function, a print statement that shouldn’t be there, a long pause, then a fix. A clean solution that arrives top to bottom at a steady typing speed, with no false starts, looks like transcription. Nobody needs an AI classifier to notice that, although HackerRank has one anyway.
does a flag fail you automatically?
Not according to anything HackerRank publishes. The help articles describe flags and reports that land in front of the recruiter, and the recruiter decides. In practice that means two different employers can treat the same flag very differently (one shrugs at a 20-second tab switch, another rejects on it). We couldn’t find a public statement of any automatic-fail threshold, so we won’t invent one.
Two caveats. The 93% figure is HackerRank’s number about HackerRank’s product, with no published methodology, so treat it as marketing until someone independent tests it. And the post is from 2023, which is a long time ago in this space. We’d guess the model has changed since, but we can’t tell you how.
The four integrity modes, and what each one can see
HackerRank’s Test Integrity help article lays out the stack. We’ve summarized it below in HackerRank’s own terms where possible:
| Mode | What HackerRank says it does | Where it looks |
|---|---|---|
| Secure Mode | Runs the test in a controlled browser environment, enforces full screen, blocks copy and paste, prevents multiple monitors, alerts on tab switches | The browser |
| Proctor Mode (AI add-on) | Everything in Secure Mode plus screenshot analysis, plagiarism detection and webcam anomaly detection | Browser, screenshots, webcam |
| Desktop App Mode (AI add-on) | Everything in Proctor Mode plus operating-system-level monitoring; the candidate installs a native HackerRank app | Your whole machine |
| Individual settings | Tab proctoring (“how often and for how long a candidate leaves the test window”), copy/paste tracking, image proctoring (periodic webcam snapshots) | Browser, webcam |
That table is the real answer to “can HackerRank detect AI?” Everything else is detail.
Browser modes vs the desktop app: why the gap is huge
A web page can only see what the browser tells it. The Page Visibility API lets a page know when its tab is hidden, focus and blur events tell it when the window loses focus, and the paste event tells it when something lands in the editor from the clipboard. That’s roughly the toolkit behind tab proctoring and copy/paste tracking. None of it lets a web page list the other apps running on your laptop.
So a browser-only HackerRank test sees symptoms. You clicked away for 40 seconds. You pasted 31 lines at once. Your solution appeared in two big bursts instead of growing the way code usually grows (write a bit, run it, fix the off-by-one, run it again). Without the webcam on, it can’t see a second device, and it can’t see a window that never takes focus from the test.
Desktop App Mode is different in kind, not degree. HackerRank describes it as “operating system-level monitoring to create a fully locked-down environment.” The help article doesn’t list exactly what that monitoring checks (running processes? window lists? screen capture?), and we won’t guess at specifics. The safe assumption is simple: if a test asks you to install HackerRank’s native app, assume it can see what else is running on that machine, including any overlay app.
We’d go further. If you get a Desktop App Mode test, that’s the employer telling you, as plainly as software can, that outside help is off the table.
What about screenshot analysis and the webcam?
Proctor Mode adds “screenshot analysis” and webcam anomaly detection. Image proctoring, per the same help article, “requires candidates to enable their webcam before starting the test” and stores periodic snapshots in the candidate report. HackerRank’s 2026 release notes fill in more than the help article does. The July 2026 notes say screenshot analysis “detects a broader range of suspicious applications, including newer AI tools, during Proctor Mode assessments,” and that object detection more accurately flags phones and tablets in the webcam feed. The April 2026 notes add that session replay now captures the full browser screen instead of a single tab, that the editor is analyzed for messages typed and then deleted (a sign of someone chatting with a helper), and that the Desktop App can detect screen mirroring.
So HackerRank is, by its own description, looking at screenshots for AI apps by name and shape. That’s a much more direct kind of detection than tab-switch counting, and it’s new this year.
What the notes don’t say is how those screenshots are captured. On Windows, an app can mark a window with WDA_EXCLUDEFROMCAPTURE (supported from Windows 10 version 2004), which Microsoft describes as a window that is “displayed only on a monitor. Everywhere else, the window does not appear at all.” macOS has a similar per-window sharingType setting. That’s how screen-share-invisible overlays work, Craqly’s included. Microsoft’s own page adds that this is not a security feature or DRM and there’s “no guarantee” it protects content, for example from someone photographing the screen.
So would an excluded window show up in HackerRank’s screenshots? If they’re taken through the standard OS capture path, probably not. If the desktop app inspects running processes or uses something lower level, it doesn’t need the screenshot at all. We don’t know which, and anyone who tells you they know for certain is selling something. The webcam is a separate problem entirely: it now looks for phones and second people, and eyes that keep sliding to the same corner of the screen look exactly like what they are.
AI-assisted tests flip the whole question
This is the bit that gets skipped in a lot of “can HackerRank detect ChatGPT” articles. HackerRank now lets employers allow AI on purpose. Its AI-Assisted Tests article describes three settings per section: No AI assistant, Guarded AI assistant, and Unguarded AI assistant. When AI is on, the report includes an “AI Assistant Transcript” with “the full conversation between the candidate and the AI assistant,” switched on by default.
In those tests, nothing needs detecting. Your AI use is the thing being graded: what you asked, whether you checked the answer, whether you understood what came back. Using an outside tool instead of the built-in one would be a strange choice, since the built-in one is the one the reviewer can actually credit you for.
HackerRank has also built an AI solvability indicator that flags library questions ChatGPT-style tools can solve, so recruiters can filter them out. That’s prevention rather than detection, and honestly it’s the smarter move on their side. A question an LLM can’t one-shot is worth more than any amount of proctoring.
A quick decision table before your test
Read the invite email and the test’s opening screen. They usually tell you which world you’re in.
- “AI assistant available” in the instructions: use the built-in assistant, openly. That’s the test.
- Full screen enforced, copy/paste blocked: Secure Mode. Tab switches and pastes are logged.
- Webcam permission requested: Proctor Mode or image proctoring. Assume snapshots of your face are reviewed.
- “Download the HackerRank Desktop App”: OS-level monitoring. Treat the machine as fully observed.
- No instructions at all: ask the recruiter. A one-line email (“Is AI assistance permitted in this assessment?”) costs nothing.
What we’d actually recommend is boring. Use AI heavily before the test (practice the patterns, have it critique your solutions, time yourself), then sit the test the way the employer set it up. If you want a real-time assistant for the live, spoken rounds where it’s allowed or for practice, Craqly gives you 15 free minutes, no card, on Mac and Windows. It’s built for Zoom, Meet and Teams calls, not for beating a proctored HackerRank assessment, and we’d rather say that plainly here than have you find out the hard way.
Where this leaves you
For the live-interview side of the same question, our breakdowns of how screen-share detection works and whether Interview Coder is detectable go deeper on overlays. If you’re weighing whether using any of this crosses a line, our honest take on AI interview assistants and cheating doesn’t let us off the hook either.
One opinion to close on, and it could be wrong: within a couple of years, the “can it detect AI” question for coding assessments will mostly disappear, because the tests that survive will be the ones where AI is allowed and watched. HackerRank already built the plumbing for that. The question that’s left is whether you can do the work with the assistant switched on.