A talent leader described a new failure mode last quarter. A senior backend candidate aced a three hour remote interview. Crisp answers, clean code, articulate system design. The team made the offer. Three weeks into the job, the engineer could not solve problems that were easier than what he had done in the loop.
The interview had been AI assisted. The candidate had a second screen, a third party feeding suggestions, and a teleprompter running in the background. The interview measured the AI, not the candidate.
This is now widespread. The two main fraud types are fake candidates posing as someone else and real candidates being coached in real time. The second is harder to detect.
What works for detection. Live video presence verification across multiple rounds catches the fake candidate case. The same person needs to show up in interview one, interview three, and the final loop. Voice biometrics work where video does not.
For real time coaching the signal is subtler. Look for delay patterns. Coached candidates have a characteristic pause before answering. They glance at a second screen. Their confidence on novel followups drops sharply because the coach cannot keep up.
The best defense is interview design, not detection technology. Open ended dialogue with rapid followups makes coaching nearly impossible. Closed problems with clean answers are trivially coachable. Most companies still rely on closed problems.
Platforms like Mazle help by surfacing inconsistencies across rounds. If a candidate's depth in round one does not match their depth in round three, the system flags it before the offer goes out.
The candidate using AI is not your problem. The interview format that makes it work is.
