1 in 3 Candidates Flagged: The 2026 Interview Fraud Playbook for Hiring Managers (Hidden AI, Proxy Candidates & What Actually Works)

1. Hidden AI assistance. An invisible overlay or second device feeds the candidate answers in real time. It's the most common form and the hardest to spot by eye.
2. Proxy candidates. Someone more skilled sits the interview, or feeds answers through an earpiece. The person who turns up on day one isn't the person you interviewed.
3. Plagiarised take-homes. Take-home assignments are now largely unreliable as a signal unless they're followed by a live walkthrough.

Why detection alone is losing

Most teams' first instinct is to buy a detector. Detection helps, but it has two problems. Cheating tools evolve faster than detectors. And text-based AI detectors produce false positives: one report found they wrongly flag a majority of writing by non-native English speakers, which risks rejecting honest candidates. The teams doing best in 2026 combine detection with interview design that makes cheating pointless.

The 2026 playbook: 6 changes that work

1. Replace silent coding with spoken reasoning. Ask candidates to explain their approach before and during the solution, with follow-up questions that change the problem ("now the input doesn't fit in memory"). Hidden AI struggles to keep up with a live, shifting conversation.
2. Look at the process, not just the answer. A keystroke replay shows how a solution was built. Large blocks of code appearing at once, with no edits, look very different from someone working a problem through.
3. Track environment signals, and treat them as evidence, not verdicts. Tab switches, window focus loss, external pastes and multiple faces on camera are useful flags. Review them alongside the answers before rejecting anyone.
4. Add an identity check at every live stage. Compare the person on camera at the first screen, the onsite and day one. A simple ID-to-camera check at the start of each live round stops most proxy attempts.
5. Decide your AI policy, and tell candidates. If AI is allowed in a round, say so and score how they use it. If it isn't, say so clearly. Ambiguity punishes honest candidates and rewards people who guess you won't notice.
6. Score against a written rubric. A consistent rubric makes anomalies stand out: a candidate who scores 10/10 on code but can't explain a single line of it is a pattern worth checking.

Red flags to watch for in a live technical screen

None of these prove fraud on their own, but two or three together are worth a closer look:

• Eye movement to a fixed point off-screen before each answer, especially after a pause.
• A delay, then a complete, polished answer to a question that would normally be worked through step by step.
• Strong code, weak explanation. The candidate can't say why they chose a data structure they just used.
• Answers that don't match the follow-up. You change a constraint and the candidate restates the original solution.
• Audio or lip-sync mismatches, or a candidate who refuses to turn the camera on for "technical reasons" at every stage.
• A resume and an interview that tell different stories, for example deep experience with a technology the candidate can't discuss beyond surface level.

Why it matters more for engineering hires

A fraudulent hire in engineering is rarely caught in week one. Code reviews, onboarding and early tickets can hide a skills gap for weeks. By the time it surfaces, the team has spent recruiter time, interview hours from senior engineers, onboarding effort and often salary, and the role has to be reopened from scratch. For a small team, one bad hire can delay a roadmap by a quarter. For remote roles with system access, a proxy hire is also a security risk: you have given production credentials to someone whose identity you never verified.

How to handle a flagged candidate fairly

False positives are real, and wrongly rejecting a strong, honest candidate is expensive too. A fair process looks like this:

1. Review the evidence, not the score. Watch the replay and look at what triggered each flag. A single tab switch to read the problem statement again isn't cheating.
2. Offer a short live follow-up. Ask the candidate to walk through their own solution and extend it. Someone who wrote the code can do this easily; someone who didn't usually can't.
3. Be consistent. Apply the same review steps to every flagged candidate, and write down what you decided and why.
4. Don't rely on AI-text detectors alone. They are especially unreliable for non-native English speakers, and a rejection based only on a detector score is hard to defend.

Where to start this week

You don't need to rebuild your whole process. Start with the stage that receives the most volume and gets the least scrutiny, which for most teams is the first technical screen:

• Triage incoming resumes against the role so only plausible candidates reach a screen.
• Move the first screen to a proctored format with spoken follow-ups and a replay.
• Have engineers review flagged sessions only, instead of running every first call themselves.