Interview Fraud and Proxy Candidates: How to Verify Who You Are Interviewing

Interview fraud is when a candidate misrepresents who they are during a remote hiring process, using a deepfake video, a stand-in who takes the interview for them, or hidden AI help to answer questions they could not answer alone. The defense that works is not one clever detector: it is verifying identity early, running a live and unscripted interview that is hard to outsource in real time, and keeping a human in the loop who can act on a bad signal. This piece explains the three forms it takes, why remote hiring made it common, and the practical steps a US hiring team can use to make sure the person you interview is the person you hire.

The problem is no longer fringe, and there are now numbers behind it. Gartner predicts that by 2028, 1 in 4 candidate profiles worldwide will be fake, a forecast backed by a second-quarter 2025 Gartner survey of 3,000 job candidates in which 6 percent admitted to taking part in interview fraud, either posing as someone else or having someone else pose as them. Experian's 2026 Future of Fraud Forecast, published January 13, 2026, listed "deepfakes outsmart HR" among its five top fraud threats for the year, warning that generative AI now produces hyper-tailored resumes and deepfake candidates capable of passing interviews in real time. Enforcement has caught up too: in a June 2025 coordinated action, the US Justice Department charged a scheme that compromised the identities of more than 80 US persons to obtain remote jobs at more than 100 US companies, with the FBI searching 29 suspected laptop farms across 16 states. Beyond the headlines, the same generative tools that write cover letters now let a moderately skilled person present as someone more qualified, or as someone else entirely. If your final round used to be an in-person meeting, that friction quietly disappeared when hiring went remote, and with it went the simplest identity check there was.

What is interview fraud?

Interview fraud covers any deliberate misrepresentation of identity or ability designed to pass a screen the candidate would otherwise fail. It shows up in three main forms, and they are not equally hard to catch.

• Proxy interviews. One person interviews, a different person shows up to do the job. A stronger stand-in, sometimes paid, takes the live interview using the applicant's name and resume. This is among the fastest-growing forms because it needs no special technology, just a second person and a shared screen.
• Deepfake candidates. Synthetic video or voice makes one person appear as another on camera, or masks a face entirely. The tooling has moved from research demos to consumer apps, which is why it made the 2026 fraud forecasts at all.
• AI-assisted answers. The real candidate shows up, but a second monitor runs a model that feeds them answers in real time. Common in technical and case-style screens, where a fed answer can lift someone a full seniority level above where they would land unassisted.

The motive is the same in all three: get past the screen. What differs is the signal that catches each one, which is why a single check rarely covers all of them.

Why remote hiring made this worse

Fully remote hiring stripped out the in-person signal that used to make fraud expensive. A candidate walking into an office for a final round is a hard friction point for a proxy, and that round has largely stopped existing for remote-first engineering, data, and support roles. At the same time, generative AI lowered the skill a fraudster needs. Passing a live coding round with a coding assistant open in a second window is now within reach of someone who could not clear it alone.

There is also a volume effect. When applying is free and instant, a single funnel can take thousands of applicants, and a small fraction of bad actors is still a lot of people to catch by eye. Cutting that pile down with software that reads every inbound resume against your criteria at least means the identity checks below run on a shortlist rather than on everyone. The honest reading of 2026 is that identity has to be treated as something you verify, not something you assume, and the verification has to sit at the front of the funnel rather than in a panicked final call.

The forms of interview fraud, and what catches each

Notice that no row is solved by a fancier resume parser. Fraud is a live-interaction problem, so the checks that work happen during a real exchange, not during document review.

How to verify who you are actually interviewing

You do not need a security team to close most of the gap. Five practices, in roughly the order they should sit in your funnel, cover the common cases.

• Verify identity at the start, not the end. Move any government-ID or verification step to the point of application or first screen, before you have sunk time into a candidate. Catching a mismatch on day one is cheap; catching it in a final round is expensive and awkward.
• Keep the same face across rounds. A proxy often taps out after the easy first call. Requiring camera-on and comparing the person across every round, ideally against the verified ID, breaks the hand-off.
• Use live challenges. Ask the candidate to do something a pre-recorded or synthetic feed cannot: hold up their ID next to their face, describe a change you just made on screen, or turn and describe their room. These are trivial for a real person and hard for a deepfake pipeline running in real time.
• Ask unscripted follow-ups. A fed answer collapses under a sharp, specific follow-up. Instead of the next scripted question, probe the answer they just gave: why that approach, what would break it, what they would do differently. Real experience survives this; a real-time assist usually does not.
• Interrogate the resume specifically. Ask for concrete, checkable detail about claimed work: the metric they moved, the tool version they used, the person they reported to. Fabrication frays fastest under specificity.

None of these require accusing anyone. They are simply the structure of a rigorous interview, and a rigorous interview is a bad environment for fraud by design.

Where an AI interview agent helps, and where it does not

An AI video interview is not identity verification, and any vendor claiming it detects deepfakes with certainty is overselling. What a well-built AI interview agent does contribute is structure at scale. It runs the same rigorous, unscripted first-round screen for every applicant, asks live follow-ups that a fed answer struggles with, and keeps a full transcript so a recruiter can spot the tells that matter: answers that do not connect, detail that evaporates under a follow-up, timing that feels off.

That structure is most valuable at the top of a high-volume funnel, where a human cannot personally pressure-test a thousand people. The agent conducts the consistent first-round screening and hands a recruiter a ranked shortlist with transcripts, and the human, who is still the one making every advance-or-reject call, applies judgment to the handful that look strong. Fraud detection stays a human decision supported by a consistent process, which is also the only version that survives an employment-law challenge.

How do you detect a deepfake in a job interview?

The most reliable low-tech method is a live challenge the candidate cannot pre-plan: ask them to hold their ID beside their face, turn their head slowly, or react to something you change on screen mid-call. Real-time synthetic video struggles with sudden, unscripted physical requests. Camera-on across every round and a face that must match a verified ID close most of the remaining gap without any specialist tooling.

What is a proxy interview?

A proxy interview is when one person sits the interview and a different person is the one who would take the job. A stronger stand-in, sometimes hired for the purpose, uses the applicant's name and resume to pass the screen. It is caught by verifying identity before the first interview and by requiring the same, camera-on person across every round, so the capable stand-in cannot quietly hand off to the real, weaker applicant later.

Can AI interviews be faked or cheated?

Any interview can be gamed, but a live, unscripted AI interview is harder to cheat than a one-way recording or a text questionnaire, because there is no time to draft or fetch an answer and the follow-ups are not known in advance. The weakest formats for fraud are ones the candidate completes alone at their own pace. Pairing a live spoken interview with identity verification at the start of the funnel is the practical way to make cheating expensive enough not to be worth it.

The takeaway

Interview fraud grew because remote hiring removed the cheapest identity check there was and generative AI made impersonation easy. You get it back not with a single detector but with a process: verify identity early, keep the same face across every round, run a live and unscripted interview that resists real-time help, and keep a human deciding. Treat identity as something to confirm rather than assume, and most of the risk falls away, without turning your hiring process into an interrogation.