Your Best Candidate Might Not Be Real. Here's How TA Catches It.

TL;DR: Over a third of live interviews are now flagged for AI-cheating behaviour, and roughly one in three hiring managers has personally uncovered a fake identity or proxy candidate. Human deepfake detection sits at barely better than a coin flip, so this isn't a vigilance problem, it's a process design problem. Verification needs to move earlier in the funnel, and TA, not IT, owns the judgment calls about where that friction goes.

‍ ‍

Let me describe a hiring process that most TA leaders would recognise as perfectly normal. ‍

A CV arrives and It's good. The candidate performs well on a screening call, articulate, well-prepared, good eye contact on video. They move through two more rounds and someone on the panel says "best candidate we've seen all quarter." An offer goes out.

Then, sometime after they start, or sometimes never, someone notices that the person who showed up for day one doesn't quite match the person who did the interviews. Different voice. Different face, slightly. A story that doesn't hold together the second time you ask about it. ‍

That is not a hypothetical anymore. That is a Tuesday.

Across nearly 19,400 live interviews analysed between July 2025 and January 2026, over a third were flagged for AI-cheating behaviour, and the rate tripled in a single three-month window. According to The Interview Guys, roughly six in ten hiring managers now suspect candidates of using AI to misrepresent themselves during the process, and one in three say they've personally uncovered a fake identity or a proxy candidate mid-interview.

This isn't confined to junior or high-volume roles either. A background screening firm found that four in ten IT, cybersecurity, risk and fraud leaders had confirmed their own organisation had hired and onboarded a fraudulent candidate and when a fraud specialist firm posted a single developer role as a test, roughly one in eight of the applications that came back were fake, including a candidate whose face was stitched together in real time by deepfake software and passed an entire panel round undetected.‍ ‍

Gartner's own projection is that by 2028, one in four candidate profiles globally could be synthetic in some meaningful way. That's not a distant number. That's roughly six recruitment cycles away for most TA teams.

Every TA leader I speak to believes that they'd spot it. They wouldn't. Nobody would, or at least not reliably.

A meta-analysis of dozens of studies on human deepfake detection found people identify a deepfake correctly at roughly the accuracy of a coin flip. The old advice, ask the candidate to turn their head quickly, wave a hand in front of their face, is now outdated as modern face-swap tools handle both of those without a visible glitch.

That should unsettle you more than it probably does. Most TA process design over the last decade has assumed that a face on a screen and a voice on a call are reliable proxies for a real, consistent human being. That assumption has quietly stopped being true, and almost nobody has redesigned the process around its absence.

Candidates cheating in interviews is not, on its own, a moral collapse. People have always polished CVs, rehearsed answers, and put their best foot forward. What's changed is the scale and the deception, not the instinct. Nearly two-thirds of hiring managers now believe job seekers are better at using AI to game interviews than TA is at catching them. That is an arms race TA is currently losing, and losing because admitting it in a board update sounds like an admission of incompetence rather than a description of an industry-wide shift.

The organisations doing something serious about it aren't the ones publicly panicking. They're the ones who've accepted that the old signals, a good CV, a confident call, a plausible LinkedIn profile, no longer mean what they used to, and have built new friction into the process deliberately.

In our new world, verification needs to move earlier, not later. Most organisations still verify identity at offer stage or after, when the fraud has already extracted maximum value from the process, weeks of interviewer time, a displaced genuine candidate, sometimes a start date. Moving identity confirmation to application or pre-screen closes that window considerably.

Live, hard-to-fake friction matters more than clever questions. Asking for a trade-off rather than a summary works, because AI-generated answers tend to be strong on structure and weak on specific, self-critical detail, the kind of thing a person who actually did the work volunteers unprompted. A candidate relaying on an earpiece tends to produce a distinctive dead pause followed by a suspiciously complete answer. That pattern is more reliable than any single gotcha question.

Some organisations are bringing back in-person interviews for final rounds specifically because of this problem, a genuinely interesting reversal after a decade of TA proudly stripping friction out of the process in the name of candidate experience. That's not a nostalgic, wasn’t it great when we met people, it's a rational one, given what remote-only verification currently can't reliably do. ‍

Here's where I want to push back on the instinct I see in a lot of organisations right now, which is to hand this whole issue to IT security or a vendor and consider it managed.

Sure, fraud detection tooling has a role, but the decision about where in the funnel friction gets added, how much candidate experience gets traded for verification confidence, and what a false accusation of fraud does to a genuine candidate's experience of your brand, those are TA judgment calls. They sit exactly in the space TA has spent a decade arguing it should own: the design of the candidate journey.

Get this wrong in one direction and you onboard fraud. Get it wrong in the other and you accuse a nervous, genuine candidate of being a deepfake because they looked at their notes off-camera. Both failures are now live risks in a way they weren't three years ago, and both sit on TA's desk whether TA wants them there or not.

In the United States, the FBI's cybercrime unit logged nearly 700 AI-related employment fraud complaints in a single year. Truffle Job scam losses reported to the FTC climbed from around $90 million in 2020 to over half a billion dollars by 2024, a trajectory that hasn't reversed. There is now a functioning black market in synthetic identity kits built specifically to help candidates pass background checks, sold to people with no technical skill required to use them.

Let’s be clear though, this isn't a candidate quietly using ChatGPT to tighten up a cover letter. This is organised, commercialised, and improving faster than most TA functions' ability to detect it.

The teams that get ahead of this aren't the ones with the best detection software. They're the ones whose TA leader took it seriously eighteen months before their peers did, and rebuilt the process around a simple, unglamorous truth: the person on the call might not be the person who shows up.

Most TA leaders are still designing their funnel as if that were still a fringe concern.

It isn't.

The question is how many fraudulent hires it takes before you do something about it.

About
Martin Dangerfield, Talent Acquisition Strategist. Straight-talker. Community builder. Talent Strategy Director at the rec hub,
I help in-house TA teams get better with a combination of sharper strategies, smarter hiring, and no-nonsense support. Since selling my business, I have been helping to grow the rec hub in its global mission to deliver an amazing embedded recruitment solution. I've built TA teams from scratch, scaled global functions, sold a business, and consult on all things talent acquisition.

I run #truManchester and #truLeeds because I believe the best ideas don't come from panels, they come from people talking honestly in a room. I'm not here for buzzwords or silver bullets.

I'm here to help TA people do the job properly.

South African-born, UK-based, and European in perspective, I'm Gen X, proud of it, and not afraid to say what others won't.

This is the work I care about. If you care too, stick around.