AI interview cheating in 2026: deepfakes, real-time coaching, and what to check for without a security team

Watch the 75-second walkthrough of Truffle’s approach. Instead of trying to detect AI, the assessment removes the right answer entirely.

You’re on a video call with a candidate for a role you need filled this month. There’s no recruiter next to you and no security team monitoring the call. You ask a question, there’s a beat of silence, and then a perfectly structured answer arrives, delivered a little too smoothly for someone thinking on the spot. Maybe that’s nothing. Maybe it’s someone reading off a second screen, or someone else entirely.

That instinct is worth trusting more in 2026 than it was even a year ago. The tools candidates use to cheat in interviews didn’t just get more common. They changed shape. Screen overlays are still around, but they’ve been joined by real-time audio coaching through earpieces, by deepfake video that swaps out who’s actually on camera, and by the same discreet smart glasses that are getting banned from the SAT for exam cheating. None of that requires the candidate to be a hacker. It requires a subscription and a Bluetooth connection.

If you’re screening candidates yourself, without a recruiter, an ATS, or anyone whose job is specifically to catch this, you don’t need a security briefing. You need to know what actually changed since last time you thought about this, what old advice stopped working, and what you can realistically check for in a normal interview.

What is AI interview cheating, and how did it expand

AI interview cheating started as a narrow problem: a candidate runs a tool that listens to your questions and displays a generated answer somewhere you can’t see, usually a hidden overlay or a second monitor. That’s still happening. But two things have widened the category since last year. (For a broader look at how AI is reshaping both sides of the hiring table, see our guide to AI in hiring.)

First, the cheating no longer needs a screen at all. Audio-only tools let a candidate sit in front of the camera with both hands visible, camera on, apparently just talking, while an AI agent listens through a hidden earpiece and feeds them the answer to repeat back. The giveaway is often timing: real people fill a pause with “let me think about that” or a half-formed sentence. Someone waiting on an earpiece tends to go silent for three to five seconds and then start a fully formed answer with no false starts.

Second, the identity of the person on the call is no longer a safe assumption. A deepfake candidate uses AI-generated or AI-assisted video, voice, or a stand-in to misrepresent who’s actually interviewing, whether that’s a fake face over a real one, a different person doing the technical round than the one who did the recruiter screen, or a proxy who’s simply more qualified than the applicant they’re standing in for.

Both of these sit on top of the original problem: a candidate quietly running a large language model to generate polished answers to your questions in real time. That’s still the most common version. It’s just no longer the only one.

How fast this is actually growing

The honest answer is: fast, but not as fast as the scariest headlines suggest, and it’s worth separating the two.

On the “candidates are actively cheating” side, Fabric, a company that runs AI-led interviews, tracked more than 50,000 candidates and found the share flagged for AI-assisted cheating behavior more than doubled in six months, from 15% in June 2025 to 35% by December 2025. That’s a vendor’s own data from its own product, so treat it as a signal from one specific interview format rather than an industry-wide rate. But the direction is consistent with everything else in this space: this is getting more common, quickly.

On the “how worried should I actually be” side, the numbers are more grounded than you’d guess. Gartner’s own 2026 survey of roughly 3,000 job seekers found only 6% admitted to engaging in interview fraud themselves. Compare that to how many hiring managers suspect it: a 2025 Checkr survey found 62% of hiring managers believe candidates are now better at faking their way through with AI than recruiters are at catching it, and 35% said they’d personally experienced someone else participating in a candidate’s virtual interview. Suspicion is running well ahead of confirmed cases.

Gartner’s often-cited long-range forecast, that 1 in 4 job candidate profiles globally will be fake in some way by 2028, is a projection, not a current measurement. Use it as a reason to build a durable process now, not as a claim about the person on your call today.

The tools moved past the screen overlay

The overlay tools we covered in earlier versions of this guide (Interview Coder, Cluely, Final Round AI, and similar products) are still around, built to render AI-generated answers invisibly on a candidate’s screen while they read from it on camera. We mapped the specific tools and the identity-verification products built to catch them in a companion guide to AI tools that detect dishonesty in video interviews.

What changed is how these companies present themselves. Cluely launched in April 2025 under the tagline “Cheat on Everything” and pulled 70,000 signups in its first week. By late April 2025 it had already scrubbed the explicit cheating language from its site, and by November 2025, after raising a $15 million Series A, it had repositioned entirely as a general AI meeting assistant. The product underneath didn’t change. Only the marketing did.

The bigger shift is that a screen overlay isn’t required anymore. Audio-only cheating needs nothing more than a Bluetooth earpiece paired to a phone running a chatbot in someone’s pocket. And the same category of smart glasses now making headlines for exam cheating uses the same discreet lens display and bone-conduction audio that would work just as well on a video interview. In June 2026, a National Taiwan University medical school applicant was caught and disqualified for wearing smart glasses concealed under an oversized frame during an entrance exam. In South Korea, two test-takers were caught using AI smart glasses during TOEIC exams in May 2026, the first confirmed cases of that specific method in a national language test there. The US College Board is banning smart glasses from the SAT starting March 2026. To be direct about the evidence: the confirmed 2026 cases are concentrated in exams, not job interviews. But the hardware and the tactic transfer directly.

Deepfakes: rarer than the fear, but real

A deepfake candidate is harder to build than a screen overlay, but not as hard as you’d hope. A 2025 Greenhouse survey of 4,136 hiring professionals found 31% had personally interviewed someone they suspected, or confirmed, was using deepfake technology. The FBI’s Internet Crime Complaint Center logged 691 AI-related employment fraud complaints in 2025. That’s a small number in absolute terms, which is a useful reality check.

Detecting a deepfake with the naked eye is genuinely hard. A meta-analysis of 56 studies found people spot deepfakes at roughly 55% to 56% accuracy, barely better than a coin flip. This is exactly why the old advice about waving a hand in front of your face or turning your head quickly deserves an update: modern face-swap models now handle both of those motions with minimal visible glitching. If you’ve read that tip somewhere, it’s outdated.

What to actually check for, without a security team

Detecting fake candidates is hard, but not impossible. None of this requires you to become a fraud investigator.

Ask for the trade-off, not the summary. AI-generated answers are strong on structure and weak on specific, self-critical detail. A real person who did the work usually has an opinion about what they’d do differently.

Listen for the pause pattern, not just the words. A candidate stalling for an earpiece to catch up tends to go dead silent for three to five seconds and then deliver a fully formed answer with no false start.

Ask them to reposition, not perform a trick. Waving a hand or turning your head is outdated. A more useful, low-friction version is asking them to adjust their camera angle or lighting mid-conversation.

Check for consistency, not just polish. Compare what someone says in the interview against their resume, their references, and later against how they actually onboard. Identity swaps and proxy interviews tend to break down here first.

Build friction in early, not just during the call. A one-way video interview already closes off part of this on its own since it leaves you a recording to review. Truffle’s one-way video interviews require a mic and camera check and a one-sitting policy candidates can’t leave partway through, a meaningfully harder environment to run a hidden second device against than an open-ended Zoom call. If you’re building a screening process around checks like this from the start, see what’s included on every Truffle plan.

If a response still looks off after that, a context signal can help you decide where to dig further. Truffle’s AI Check flags linguistic patterns that suggest a response may have been AI-assisted and surfaces that alongside the rest of a candidate’s record, not as a verdict, but as a reason to ask a sharper follow-up question.

Why this matters more on a small team

A bad hire costs more when there’s no bench to absorb it. A 2025 Checkr survey found 23% of companies lost more than $50,000 to hiring fraud in the past year, and 10% lost more than $100,000. On a four-person team, that’s not a line item. That’s the budget for the role, and then some.

What actually holds up is a process with enough real, personal, hard-to-fake friction built in from the start, so that gaming one step doesn’t carry a candidate through the rest of it.

FAQs about AI interview cheating

Is AI interview cheating actually common, or is this overstated?

Both things are true at once. Vendor data shows cheating attempts climbing fast, and hiring managers are increasingly suspicious. But confirmed, documented cases stay much smaller: the FBI’s IC3 logged 691 AI-related employment complaints in 2025, and only 6% of job seekers admit to interview fraud in Gartner’s own survey.

Do the old anti-deepfake tricks, like waving a hand or turning your head, still work?

Not reliably. Modern face-swap deepfakes now handle quick head turns and hand movements with minimal visible glitching. A better use of that moment is asking someone to naturally reposition their camera mid-conversation.

Can candidates cheat without any software on their screen?

Yes. A Bluetooth earpiece paired to a phone running a chatbot needs no software on the interview device at all. The clearest tell is timing: three to five seconds of silence after a question, followed by an unusually polished answer.

What should I do if I catch a candidate cheating?

Document what you saw or heard and disqualify the candidate. Then look at your own process for where it still runs on trust instead of evidence, and fold that into how you spot red flag candidates earlier next time.

Does AI interview cheating detection software actually work?

It helps, but it isn’t a verdict machine. The best tools flag patterns worth a second look so you know where to ask sharper follow-up questions. The strongest setups combine a flag from software with a human decision.

Is this only a problem for tech companies hiring for coding roles?

No, though technical interviews remain easiest to game. Behavioral rounds are harder for AI to fake convincingly, especially with situational interview questions built around a specific past decision. But identity fraud and deepfakes don’t care what the role is.

Cheating tools will keep getting better, and so will the ways to spot them. What doesn’t change is the value of a process built around specifics a real person actually lived through. See how Truffle’s plans build that in from the start.