Technical Assessment Platforms

Assessment cheating has doubled in a year — and detection is the new battleground

Source: CodeSignal / HireVue · Sep 25, 2026

The scale of assessment fraud is now measurable, and it's rising fast. CodeSignal reports that cheating and fraud on proctored technical tests roughly doubled in a single year — from 16% in 2024 to 35% in 2025, and closer to 40% at the entry level. In response it now scores every submission with a 'Suspicion Score'; among flagged tests, 35% showed off-screen referencing and 23% showed unnaturally linear typing patterns. HireVue takes a complementary route, using tab-focus logging and cross-candidate answer-similarity scoring, where a match above 0.4 flags likely script-sharing.

What unites these approaches is that they look at behavior, not just outcomes. A correct answer no longer proves competence; how the answer was produced — the timing, the attention, the similarity to other submissions — is where the real signal lives. That is a fundamentally harder thing to fake, and a fundamentally richer thing to measure.

It also points squarely at multimodal, process-level interview data. Distinguishing genuine reasoning from assisted performance means modeling what authentic problem-solving looks like as it unfolds — across input behavior, timing, audio and video — which can only be learned from large volumes of real, consented interview recordings. Detection and authentic data are two sides of the same asset.

Key Points

• CodeSignal: cheating/fraud rates on proctored technical tests doubled in a year — 16% (2024) to 35% (2025), ~40% at entry level
• It now scores every submission with a 'Suspicion Score'; of flagged tests, 35% showed off-screen referencing and 23% unnaturally linear typing
• HireVue relies on tab-focus logging and cross-candidate answer similarity (a ≥0.4 match flags script-sharing)
• The signals that matter are behavior-level, not outcome-level — how an answer was produced, not just whether it's correct