Building a fake job candidate used to take real production value: a stolen or invented identity, a rehearsed backstory, someone willing to sit through a video call and improvise convincingly under pressure. It was possible, but it took effort, and effort is a natural filter. That filter is mostly gone now.
Researchers have demonstrated that someone with zero prior experience in image manipulation can build a synthetic candidate capable of passing a live video interview in roughly 70 minutes. That's not a static photo or a doctored resume. It's a real-time deepfake, good enough to hold a conversation, react to unscripted questions, and sustain the illusion for the length of an actual interview. The tools involved aren't specialized or hard to find. They're the same category of consumer AI software that has spent the last two years getting cheaper, faster, and dramatically easier to use for people with no technical background at all.
Hiring teams have noticed. A 2026 survey found that 59 percent of hiring managers had suspected at least one candidate of using AI to misrepresent themselves during a hiring process, whether that meant a fabricated identity, a stand-in doing the actual interview, or credentials that didn't hold up under questioning. That's not a concern confined to a handful of security-obsessed companies. It's approaching a majority.
The more striking number is what happens after the interview ends. A separate study found that 41 percent of large organizations had already onboarded someone who turned out not to be who they claimed to be. That person wasn't flagged in a background check or caught in an exit interview. They were hired, given credentials, and given access, and the mismatch surfaced later, if it surfaced at all. Multiply that by however many systems a typical new hire touches in the first month, source code, customer data, internal tools, and the exposure is not a hypothetical either. It's whatever that person decided to do with the access they were given.
The people catching this are losing ground
Ask hiring professionals who's ahead in this particular contest and the answer is uncomfortable. In one survey, 62 percent admitted that job seekers are now better at faking identity with AI than HR teams are at catching it. That's a rare thing for an industry to say about itself out loud, and it's worth taking at face value rather than dismissing as excessive modesty. Detection tools exist, but they're built to catch yesterday's fakes, and the technology producing new ones moves faster than most enterprise procurement cycles.
It's also, structurally, a difficult contest to win through detection alone. Every tell a verification tool learns to spot, an unnatural blink rate, a lighting mismatch, a lag between audio and lip movement, becomes a known problem for the next version of the software generating the fake. Detection is a race against something that gets to see the finish line before it runs.
There's also a simpler reason detection struggles: most hiring teams aren't security teams. A recruiter or hiring manager running a video interview is evaluating communication skills and technical judgment, not scrutinizing pixels for compression artifacts. Asking that same person to also be the last line of defense against synthetic video is asking them to do a second job they were never trained or equipped for.
What actually holds up
The more durable answer isn't a sharper real-time detector. It's moving verification away from the interview entirely and back to the source: confirming identity against the document or database that issued it, confirming employment against the payroll or records system that actually paid the person, confirming skills against the platform or institution that certified them, rather than against how convincing someone sounds on a call. A deepfake can pass a conversation. It has a much harder time passing as a match against records it had no hand in creating. That's roughly the logic behind verification-at-source platforms, Dealt included: check a claim against where it originated, not against how well it performs live.
None of this makes the video interview useless, and it doesn't mean every hiring team needs to overhaul its process by next quarter. But the numbers say the old assumption, that a live conversation is a reasonably reliable proxy for a real person, isn't one hiring teams can take for granted by default anymore. The interview used to be the test. Increasingly, it's just the opening move.