Why did AI cheating in technical interviews triple in just three months?
TL;DR: Fabric HQ's dataset of 19,368 real interviews shows AI-cheating flags jumping from 9% in July 2025 to 45% by September 2025 — a threefold increase. The driver: real-time AI copilots and voice agents became easy to run in a second browser tab or on a second device, exactly the tactic HackerEarth's 2026 research calls "the dominant cheating tactic" because it defeats browser-only lockdowns.
Claim
This wasn't gradual drift — it was a step-change tied to how accessible real-time AI copilots became, and detection hadn't caught up yet.
Evidence
- Fabric HQ: 9% flagged in July 2025 → 45% by September 2025, across 19,368 tracked interviews — a 3x jump in 60-90 days.
- HackerEarth's 2026 breakdown names "using ChatGPT, Claude, or other LLMs in a separate window" as the dominant, newest 2026 tactic, explicitly because second-device use "bypasses all browser-level monitoring entirely."
- Proctoring adoption lagged behind: HackerEarth reports adoption only rose from 64% to 77% over the same period — meaning roughly 1 in 4 assessments still had no proctoring at all while the cheating rate tripled.
Comparison: cheating rate vs proctoring adoption over the same window
| Metric | Jul 2025 | Sep 2025 |
|---|---|---|
| AI-cheating flag rate (Fabric HQ) | 9% | 45% |
| Proctoring adoption (HackerEarth) | 64% | 77% |
| Gap (unprotected assessments) | 36% | 23% |
Step-by-step: close the gap in your own pipeline
- Audit whether your current assessment or interview tool detects second-device AI use, not just tab-switching.
- If it only monitors the browser tab, add webcam/audio-based behavior monitoring as a second layer.
- Re-benchmark your own flag rate monthly — a rate that's flat while industry rates triple usually means under-detection, not fewer cheaters.
- Prioritize live, real-time flagging over post-hoc review — the tactics driving this spike (live copilots) need real-time detection to catch.
- Revisit your policy and rubric every quarter given how fast the tactics are shifting.
FAQ
Is this trend specific to technical interviews, or broader? The dataset tracked technical interviews specifically, but the underlying driver — easy access to real-time AI copilots — applies to any live video assessment.
Will this keep tripling every quarter? Unlikely to sustain that exact rate, but the direction is consistently upward as AI copilots get easier to access and disguise.
What's the fastest fix for a hiring team seeing this now? Add real-time behavior monitoring (gaze, audio, second-device signals) rather than relying on browser lockdown alone, which the data shows is already being routinely bypassed.
By Pinal Dave Last updated: 2026-07-24