Proxy interview fraud is a continuity problem, not an ID check

Proxy interview fraud is when someone other than the real candidate sits a hiring interview for them, usually a remote technical round, so an applicant who can't do the job gets through. A single ID check at the start won't stop it, because it only proves someone appeared once. What stops it is treating every round as evidence that it's the same person.

If you landed here after typing something like "fake candidates remote interview" into a search bar following a strange call, you're not imagining things. The FBI has issued public warnings about fraudulent remote applicants, and the Justice Department has prosecuted schemes built on stolen identities.

One scope note. Our guide on how to detect AI cheating in interviews answers the question "is this the candidate's own answer?" This piece answers a different one: "is this the same person every time?"

Key Takeaways

Our position: proxy interview fraud is a continuity problem, not a checkpoint problem. Verifying identity once proves someone matching the ID showed up once, and nothing more.
The FBI warned in June 2022 that deepfakes and stolen personal data were being used to apply for remote IT, programming, database, and software jobs.
In June 2025 the Justice Department said one North Korean remote IT worker scheme, charged in Massachusetts, used more than 80 compromised US identities to get jobs at more than 100 US companies.
In a technical hire, the strongest identity signal is consistency across rounds: the same voice, the same reasoning habits, and the same depth on the candidate's own projects.
A structured first round with a full transcript gives every later round a baseline to compare against.

What proxy interview fraud actually is

A proxy interview is one where the person answering the questions isn't the person you'll end up hiring. The classic version is a bait-and-switch: a skilled stand-in clears the technical rounds, and the real applicant shows up on day one. Variants include someone feeding answers off camera, someone remote-controlling the candidate's machine, and at the far end, a synthetic face laid over another person's voice.

It's different from AI-assisted cheating, and the difference decides what you do about it. With cheating, the right person is getting outside help. With candidate impersonation, the wrong person is there entirely, so checking whether answers are original won't catch it. If your worry is copilots and pasted answers, what AI cheating in interviews looks like is the better read.

Remote technical hiring made the switch cheap. On a video call, the only thing tying the face on screen to the resume is whatever the interviewer checked, and most engineers running a first round are checking code, not identity.

There's a supply side too. Search this topic and the first page includes Reddit threads asking about proxy interviews arranged through Telegram groups and paid "job support", which tells you some of it is sold as a service rather than done as a favour.

What the FBI and DOJ have actually documented

Most pages ranking for this problem cite no primary source at all. The primary record is small but specific. We couldn't find a trustworthy figure for how often proxy interviews happen across all hiring, so we won't invent one.

On June 28, 2022, the FBI's Internet Crime Complaint Center issued a public service announcement on deepfakes and stolen PII in remote job applications. It named IT, programming, database, and software roles, and said background checks sometimes showed the personal details submitted belonged to someone else. Its video warning was specific: on-camera lip movement and actions didn't fully line up with the audio.

In January 2025, the FBI followed with a PSA on North Korean IT workers conducting data extortion. Workers hired under false identities were copying code repositories, harvesting credentials, and holding stolen data for ransom. Several of the mitigations it lists are about the hiring process: verification during interviewing, onboarding, and throughout employment, targeted questions about location and education, and in-person hiring where possible.

Then on June 30, 2025, the Justice Department announced coordinated nationwide actions against North Korean remote IT worker schemes. It reported searches of 29 known or suspected laptop farms across 16 states. In one case charged in Massachusetts, prosecutors said the scheme used the identities of more than 80 US people to get remote jobs at more than 100 US companies, costing those companies at least $3 million in legal fees, remediation, and other losses.

These are organised operations, not a friend helping a nervous graduate. And the FBI's own advice is to keep verifying through interviewing, onboarding, and employment rather than once at the door.

Why a single ID check is not enough

The standard advice is a checkpoint. Hold a government ID up to the camera, match it to the face, move on. That's worth doing, and it answers one narrow question: did a person matching this ID appear in this call?

It doesn't answer the question you care about, which is whether the person who passed the technical screen is the one in the final round and the one who logs in on day one. A proxy scheme is built around exactly that gap. The stand-in only needs to be present for the rounds that test skill, and the real applicant can handle the ID check, the culture chat, and the offer call.

Lockdown tools share the blind spot. Blocking apps and detecting second screens helps against AI copilots, but if the proxy is the one sitting at the locked-down machine, the lockdown is protecting the fraud. The machine is clean and the person is wrong.

Identity as continuity: treating every round as evidence

Here's the reframe. In a technical hire, the strongest identity signal isn't a document, it's continuity: the same voice, the same way of reasoning through a problem, and the same depth of knowledge about the candidate's own projects, from the first screen to the final round.

A stand-in can pass one round beautifully. What's hard is keeping a story straight across four rounds when two different people are taking them. The stand-in knows the system design answer, and the real candidate knows which database they migrated last year and why it hurt, and those two sets of knowledge rarely overlap cleanly.

That means your loop needs three things it probably doesn't have today:

A baseline, meaning one early round captured in full rather than summarised into a thumbs-up.
Shared threads, where later rounds deliberately revisit specifics from earlier ones.
A comparison step, where someone reads the earlier evidence before the next round, not after the offer.

None of this needs an accusation. It's what a good structured technical interview already does: build on what came before, and notice when the person in front of you doesn't remember it.

The AI first round as a baseline

This is where running the first round with an AI interviewer changes things, though not the way people assume. A proxy can sit an AI interview as easily as a human one. What changes is what the round leaves behind.

A human first round usually produces a few lines of notes and a score. An AI-run round on Expert Hire produces a scored report card with the full transcript behind it: what the candidate said about their own projects, how they talked through the live coding problem, and the code they wrote. How AI interviews are scored walks through what's in that record.

That transcript is the baseline. Before the final round, the interviewer can read how this person described their last system and then ask about it. If the answers in round four don't match the voice and depth of round one, you've learned something no ID check would have told you.

During the AI round itself, Expert Hire's desktop proctoring app (included on the Growth plan and above, and never metered against your credits) adds tab-switch awareness, screen monitoring, and voice-consistency checks during the session. None of that is face matching or deepfake detection, so the honest claim is narrower: a consistent, transcribed baseline makes a swap between rounds much harder to hide.

Depending on your plan, a recruiter can also join a live AI interview as a silent listener and take over if something feels off. For the final round a human takes the seat, which the human-led interview post covers, and the research behind our rubric design is on the methodology page.

Questions a stand-in cannot prepare for

Structure is what makes rounds comparable. Sackett and colleagues' 2022 reanalysis in the Journal of Applied Psychology put structured interviews at roughly .42 validity, revising Schmidt and Hunter's 1998 figure of .51 downward, with structured interviews still ranked above unstructured ones. For proxy detection, structure has a second benefit: the same rubric across rounds means you're comparing like with like.

The questions that expose a stand-in are the ones only the real candidate can answer, because they're tied to that person's history:

"Walk me through the last bug you shipped a fix for. What did the logs show?"
"In the first round you said you moved off your old queue. What broke during the migration?"
"Why did your team pick that database over the alternative, and would you pick it again?"
"Change one constraint on the problem you solved earlier. What would you do differently?"

Asking a candidate to extend code they wrote in an earlier round is especially hard to fake. Whoever wrote it can extend it, and someone who didn't will struggle to explain their own variable names. Make at least one question in every later round depend on an earlier round's answer.

Keep this respectful. Genuine candidates forget details and get nervous, so one fumble isn't evidence, while a pattern across rounds is. Our comparison of AI interviews and human interviews is useful context on which round should ask what.

What to do when you suspect a proxy

Suspicion isn't proof, and handling it badly can burn a real candidate. A calm sequence works better than a confrontation.

Write down what you saw, with timestamps: audio that doesn't sync with lip movement, a voice that differs from an earlier round, answers that contradict the first-round transcript.
Compare against the baseline before deciding anything. Read the earlier transcript and ask whether the gap is real or just nerves.
Ask a short follow-up on something only they would know, live with the camera on, ideally revisiting their own earlier work.
Re-verify through your normal process, and check your ATS or HR system for other applicants with the same resume content, phone number, or email, which the FBI's 2025 PSA specifically recommends.
Stop and document if the evidence holds. If the role touches sensitive systems and you suspect a state-linked scheme, the FBI takes reports through IC3.

What not to do: don't accuse on a hunch, and don't rely on a single tell. The lip-sync signal the FBI described is real, but a bad webcam and a weak connection can produce the same effect.

Frequently asked questions

What is a proxy interview?

A proxy interview is a job interview where someone other than the real applicant answers the questions, either by appearing in their place or by feeding them answers. It's most common in remote technical hiring, where a stand-in clears the skill rounds and the real applicant takes the job.

How is candidate impersonation different from AI cheating?

Cheating is the right person getting outside help, such as an AI copilot. Candidate impersonation is the wrong person being there at all. You catch impersonation by comparing rounds, not by checking whether an answer is original.

Can an ID check on camera stop interview identity fraud?

It stops the laziest version. It proves a person matching the ID appeared once, not that the same person sat every round, so treat it as one input among several rather than the answer.

How do you detect proxy interview fraud in a remote loop?

Capture one early round in full, make later rounds revisit specifics from it, and have someone compare the two before the next round. Watch for voice changes between rounds, answers that contradict earlier ones, and thin knowledge of the candidate's own projects.

Does an AI interviewer prevent proxy interviews?

No tool prevents them outright. An AI first round gives you a consistent transcript that later human rounds can check against, and Expert Hire's desktop proctoring includes voice-consistency checks during the session, but it isn't face matching or deepfake detection.

Check identity every round, not once

Proxy interview fraud survives because hiring loops verify identity once and then trust that result for weeks. The FBI and DOJ records show the identity-fraud end of it is organised and aimed squarely at remote technical roles. A stricter checkpoint won't fix that. Treating every round as evidence, and actually comparing them, will.

The simplest place to start is a first round that leaves a full record. See how the Expert Hire AI interview platform builds a scored report card with the transcript behind it, then ask yourself whether your final-round interviewer would notice if the person in front of them wasn't the one in that transcript.

By TK, Growth at Expert Hire. Last updated September 29, 2026. Reviewed by Anand Suresh, CPO at Expert Hire.