How HackerRank Stops Cheating in Tech Interviews

Key Takeaways

AI wrote the resume. Claude completed the assessment. Someone else showed up to the interview. And the actual hire, they used invisible cheating tools to get through. After a decade helping the largest companies in the world conduct tech interviews, we've seen more suspicious behavior than anyone else. So, we built a fix. Three modes, each targeting a different layer of the problem. This is on by default for every test on HackerRank. The moment a candidate starts their assessment, it opens in full screen. If they try to exit, the mode [music] catches it and pulls them back. Any tab switch or move out of full screen [music] gets tracked and shows up in the report. We recommend keeping secure mode on for every role, every geography. It's the baseline standard. But, it only covers the basics. Proctor mode goes deeper. Proctor mode is part of our AI add-on. You can enable it in your test settings. Think of it like [music] Clippy, but for ensuring fairness and way less annoying. When the test begins, the proctor lays down the rules, asks for permissions, and then [music] gets out of the way. A focused, honest candidate never sees it again. But, the moment something looks off, it steps in [music] and that moment gets logged. Every flagged event ends up in one report with an integrity [music] rating. None, medium, or high. High means clear evidence. In this example, our object detection flagged them using their phone multiple times during the test. [music] Medium means something was off, but no confirmed malpractice. None means a clean [music] session. You get a full session replay with suspicious moments already marked, so you're not making judgment calls without evidence. Every candidate gets a fair environment to show what they actually know. But, there's still room for cheating. Their room. Beyond the screen, the webcam is always on. If a candidate steps away too long, they get a warning. The system also detects phones, earphones, or any object that could give an unfair edge. It's watching the environment, not just the screen. Secure mode and proctor mode live in the browser. The browser can see a lot, but not everything. A remote desktop tool or an invisible AI overlay can't be detected by browser modes, but that's why we built the desktop app. Before the test begins, the app scans for blacklisted tools and forces them to close. If a candidate tries to reopen one mid-test, it shuts it down automatically. Invisible tools like Clooly and UltraCode are not invisible to it. >> [music] >> Copy-pasting from somewhere else is disabled entirely. Proctor mode catches suspicious behavior in real time. The HackerRank app stops it before it starts. The same proctor's still there, same interface, [music] just with deeper permissions and more visibility. Candidates see it as a signal you're running a serious process. Everything the three layers catch ends up in one place. One report per candidate with their score, their integrity rating, and a full breakdown of flagged events. And all of this is configurable. If you're heading to a campus that's to hire junior developers, you'd probably want to turn on desktop mode with image analysis. But if you're Zuckerberg hiring an AI researcher, you'd probably want to disable most of it and just pay them a hundred million dollars. We've talked a lot about proctoring and integrity, but every layer we've built keeps the candidate experience front and center. The environment mirrors what developers see at work. Real-world problems in an IDE with an AI assistant that includes agent mode and lets them choose from the latest models. They can actually showcase their skills as an orchestrator. We're committed to building the best hiring infrastructure so you can hire the right people with confidence. I look forward to seeing you on the other side of the demo.