Contract Cheating in 2026: Proxies, Remote Access, and Hidden AI

The three shapes, and which one is growing

Contract cheating means outsourcing part or all of an assessment to someone else and submitting the work as your own. It used to be low tech: pay someone to write the essay. The 2026 version has three shapes, and they need different defenses.

The first is hiring: a candidate pays a proxy to take a pre-hire assessment or sit a certification exam. Research puts the scale at about 8 to 9 percent of college students paying someone to complete an assignment, and 10.5 percent having someone else take an online test for them. The second is remote access: the candidate lets another person control their computer through remote access software and answer the questions while they stay on screen. The third is hidden AI: tools built to stay invisible even during screen sharing, feeding answers in real time.

The proxy is the easiest to catch and the hardest to prove

A proxy fails on identity, not on answers. Government ID checks, facial matching, and voice biometrics before the exam, repeated through the session, catch a substitution that a single check at the start misses. Keystroke rhythm and typing patterns add a second layer: the person typing in minute forty should type like the person who typed in minute five.

The hard part is what happens after the flag. A mismatch is a lead, not a verdict, and a wrong call ends a career or a degree. False positives are the reason every flag has to go to a trained human with the full session recording, not to an automatic fail. The Duolingo English Test runs this model at scale, analyzing patterns across thousands of sessions at once, including near-identical answers from test takers in the same region, which no single proctor watching one session could ever see.

Hidden AI is the shape with no good answer yet

The AI assistant is the hardest of the three, because it is designed for the exact situation it cheats in. Second-device chatbots, on-screen overlays, and voice clones that need a few seconds of audio all postdate the camera-on-face vendors, and the vendors are patching, not rebuilding. Honorlock's own guidance admits the gap: new AI tools are hard to detect because they are specifically built to remain hidden, even during screen sharing.

The honest answer is that detection alone will not close this one. The defenses that work are structural: assessments the AI cannot do for the candidate, like a live oral defense of the answers, a follow-up question that changes the problem, or a task that requires the candidate's own data. Take-home exams already point this way: the assessments that survive are the ones a bot cannot game, not the ones with a better lock on the browser. Contract cheating is a market. It grows where the assessment is worth cheating on and easy to cheat on. Change one of those two, and the market shrinks.