Technical Assessment Platforms

Interview cheating has become an integrity crisis — and multimodal detection is the answer

Source: HeroHunt / industry data · Sep 23, 2026

AI-assisted cheating has crossed from an edge case to a structural problem in hiring. A striking 83% of candidates told HackerRank they would use AI on an assessment if they were confident it couldn't be detected; 65% of hiring managers report already catching deceptive AI use, and roughly 22% of US job seekers admit to using AI during live interviews. Candidate-side copilots like Cluely and Interview Coder, combined with second-device setups, have made old-style proctoring easy to defeat.

The response taking shape is not more invasive lockdown software — it's smarter signal. Rather than watching a screen for forbidden behavior, the emerging approach fuses multiple modalities at once: video, audio, behavioral cues, and device telemetry, paired with adaptive questioning that adjusts in real time to how a candidate actually reasons. The bet is that authenticity is far harder to fake across several correlated channels than on any single one.

This is the center of the interview-integrity problem, and it maps directly onto why genuine, multimodal interview data matters. Detecting whether a response is real requires understanding what real looks like across voice, expression, timing and reasoning together — which in turn requires large volumes of authentic, consented, multimodal interview recordings to learn from. The cheating arms race is, underneath, a data problem.

Key Points

  • HackerRank: 83% of candidates say they'd use AI on an assessment if they were sure it was undetectable
  • 65% of hiring managers report catching deceptive AI use (Greenhouse); ~22% of US job seekers already use AI in live interviews (Resume Genius)
  • Candidate-side tools (Cluely, Interview Coder) plus secondary-device stealth are outrunning traditional proctoring
  • The industry is shifting toward multimodal detection — video + audio + behavioral + device signals — and adaptive questioning, not lockdown surveillance