Fake Candidates Get Through Where Nobody Talks to Them

Candidate fraud used to be a story recruiters swapped at conferences. Someone knew someone whose new hire showed up on day one looking nothing like the person from the video interview. It now has a Gartner number attached: in a 2025 survey, 6% of candidates admitted to some form of interview fraud, and Gartner expects that by 2028 one in four candidate profiles worldwide will be fake [1].

Most of the advice I've read since then is about detection. Deepfake detectors, liveness checks, proctoring software that tracks where a candidate's eyes go. Some of that is worth doing, but it misses where the problem comes from. Fake candidates rarely beat a good conversation. They get through processes that have very few conversations in them, and over the last five years most of us built exactly that kind of process.

How the funnel got quiet

Walk through a typical remote hire for a mid-level engineering or support role. The candidate applies through a portal. A parser scores the resume. If it clears, an email invites them to a recorded video screen or a timed assessment. Then a quick recruiter call, then a panel. The first time anyone asks the candidate a follow-up question they couldn't have seen coming is often round three.

Every one of those steps was added for a decent reason, and the reason was usually volume. Recruiters couldn't phone 400 applicants per req, so we swapped the phone call for things that scale: forms, recorded answers, take-home tests. What those have in common is that the candidate prepares alone, and nobody checks who is doing the preparing. For someone running a fraud operation, it's close to the best setup you could ask for.

The organized version of this is not small. In mid-2025 the US Justice Department announced coordinated actions against North Korean remote IT worker schemes, including searches of 29 suspected "laptop farms" across 16 states [2]. I'd bet very few of those workers got hired on the strength of a long, unscripted conversation early on.

Why checking harder at the end doesn't fix it

The instinct is to bolt verification onto the back of the funnel: an ID check before the offer, a stricter background check, maybe one mandatory onsite. Do those things, but look at the cost. By the time you're checking a passport, a recruiter, a hiring manager and three panelists have already spent hours on this person. If they turn out to be fake, those hours are gone, and the real candidates queued behind them have been waiting.

Detection also puts you in an arms race you will sometimes lose, because face-swap and voice-cloning tools get better every few months. And the surveillance approach lands hardest on honest people. Asking every applicant to install proctoring software tells good candidates you assume they're lying. Some of them leave, and they tend to be the ones with other offers. We've written before about how recorded screens end up rewarding compliance over ability, and heavier monitoring pushes further in that direction.

The cheaper fix sits earlier. Put a real conversation near the top of the funnel, and make it one that's hard to fake.

What a hard-to-fake conversation sounds like

What trips up a fraudulent candidate usually isn't a trick question. It's an ordinary follow-up. You said you moved the billing service to Postgres, so what broke first? Who did you report to on that team, and how many people were on it? Why did you leave in March instead of after the launch? Someone who did the work answers without much thought. Someone reading from a script, or relaying answers through a second person, gets slower and vaguer with each question, and the details stop lining up with the resume.

That is what a good phone screen always did. It also takes a recruiter's full attention and doesn't stretch to 400 people a req. So most teams dropped it. I'd argue some of today's fraud problem is the bill for that decision, the same way a lot of talent shortages turn out to be screening shortages.

Where Asendia AI fits

This is the part about what we built, so weigh it accordingly. Asendia is a voice-first AI recruiter. It phones every applicant, usually within hours, 24/7, and runs a structured screening conversation. It asks your recruiters' questions and follows up on what the candidate actually said, so the answers can't be polished in advance the way a recorded video answer can. The transcript and a structured summary land in your existing ATS next to the application.

I want to be careful about the claim here. Asendia is not a deepfake detector, and I wouldn't tell anyone to drop identity verification because they use us. What it changes is where the first real conversation happens. When a candidate's dates shift between the resume and the call, or they can't say one specific thing about a project they listed, that's in the transcript before a hiring manager has blocked an hour for them. Recruiters can then spend their hours on verification and on the candidates who are clearly real.

Most staffing agencies using Asendia came for volume, screening hundreds of applicants a week without adding recruiters. But once every applicant has had an actual conversation, the odd ones stand out much earlier.

Final Word

Some of the money going into fraud detection will be well spent. I'd still start with a less exciting question: at which step does a candidate first have to answer something they didn't see coming? If the honest answer is round three, that's where your exposure comes from, and a scanner at the end won't move it.

Pull your last ten remote hires and check when each of them first had an unscripted conversation with someone at your company. If the answer is "the final panel" for more than a couple of them, fix that before you buy anything.

Ready to transform your hiring strategy? Schedule a Demo with our founders today!: https://asendia.ai/talk-to-founders