How deepfakes and proxy candidates beat traditional screening

The candidate on the video call answered every question well. The resume was clean, the references checked out, and the interview panel agreed she was the strongest applicant they had seen for the role. Three weeks after the offer, the work that showed up did not match the person who had interviewed. By then, the mis-hire had already cost the team a quarter.

Stories like this are becoming more common, and the reason is not that recruiters are getting worse at their jobs. It is that the two things most hiring processes rely on most heavily, the resume and the live conversation, have become the easiest parts of hiring to fake. Generative tools can write a convincing resume in seconds, coach a candidate through an interview in real time, or place a synthetic face and voice on a video call. Gartner has projected that by 2028, one in four candidate profiles worldwide could be fake.

Here is the argument this article makes: candidate fraud is a screening-method problem, not a detection arms race. You will not win by buying more detectors to catch better fakes. You win by shifting the weight of your decision onto evidence that is genuinely hard to fake, which is observed performance on realistic, job-related work. Detection still has a role. It is just the wrong foundation to build a hiring decision on.

What candidate fraud looks like in 2026

Candidate fraud is the deliberate misrepresentation of who a candidate is or what they can do, in order to pass a hiring process. It is not the same as a candidate using AI to sharpen an application, which is a question of where you draw the line rather than an integrity breach. That distinction between augmentation and fraud is worth setting out clearly before you police it, as our own Kasey Harboe Guentert argues. Fraud is the harder end of the spectrum, and it now shows up in a few recognizable forms.

Deepfake candidates use synthetic audio or video to disguise identity during interviews, so the person on screen is not the person who would do the job. Proxy interviews put a more qualified stand-in on the call, or feed a candidate answers through a second device, so the assessed skill belongs to someone else. Identity fraud goes further, using stolen or fabricated identities to get hired at all. The most organized version of this is now a documented national security concern: in 2025, the U.S. Department of Justice and FBI took coordinated action against North Korean IT workers who used stolen and fake identities and third-party proxies to secure remote roles at American companies.

The proxy-interview tactic is not hypothetical at scale. In the North Korean case, the FBI documented U.S.-based facilitators who attended virtual interviews impersonating the actual workers, so the person assessed on the call was never the person who would hold the job.

What these tactics share is that they all target the same two checkpoints. They defeat the resume, and they defeat the conversation. Any hiring process that leans mostly on those two inputs is exposed, no matter how experienced the interviewer is.

Why "detect the fake" is the wrong foundation

The instinct, when fakes get better, is to get better at spotting them. Add identity verification. Add a liveness check on the webcam. Add software that flags a second screen or an unusual typing pattern. Each of these can help, and enterprise hiring at scale does need reasonable integrity controls. The same logic applies inside the interview itself: knowing the classic tells of a lying candidate is still worth building into a structured process, even though it isn't the foundation the hiring decision should rest on.

The problem is structural. Detection is a reactive posture. Every new control teaches the next generation of fraud what to avoid, and the tools doing the faking improve faster and cheaper than the tools doing the catching. You end up in an arms race you have to keep funding, where success is defined as not losing, and where a single missed fake still lands a bad hire on a team.

There is a deeper issue too. Detection answers the question "is this person real?" It does not answer the question that actually determines the outcome of the hire: "can this person do the work?" A candidate can be entirely who they say they are and still be wrong for the role. So even a perfect fraud detector leaves the core hiring risk untouched. Identity verification in hiring is a floor, not a decision.

The same shift plays out on the assessment side of hiring. Job simulations resist cheating by changing what a candidate has to do, not by watching them more closely -- see why job simulations resist cheating by design.

A more useful way to think about it: the fakeability of evidence

The more durable frame is to sort your hiring evidence by how hard it is to fake, and then decide how much weight to place on each type.

A useful test for any piece of evidence is simple: how much genuine, role-relevant skill does a candidate need in order to produce it? The less skill required to fake it, the less it should decide.

• Resume and credentials — easy to fake: self-reported, unverified, and now generatable in seconds.
• Live interview answers — easier to fake than it used to be: coachable in real time, and vulnerable to proxies and deepfakes.
• Observed performance on realistic tasks — hard to fake: faking it requires actually doing the work to the required standard.

That last row is the point. When you ask a candidate to complete a realistic, job-related task and you watch how they approach it, the only reliable way to "cheat" is to genuinely perform at the level the role requires. If someone can consistently do the actual work well, the label you put on that, a proxy or a candidate, matters far less. The evidence is the skill itself, demonstrated, rather than a claim about the skill.

This does not make performance-based evidence tamper-proof, and it does not remove the need for sensible integrity controls or human judgment. It changes what your decision rests on. A hiring process weighted toward demonstrated performance is expensive to defraud, because the cost of faking it is doing the job.

Why this matters more at enterprise scale

For an enterprise talent acquisition function, the fraud problem is not a handful of suspicious candidates. It is a volume and consistency problem. When you are screening thousands of applicants across regions and roles, you cannot rely on a single sharp interviewer to catch an anomaly. You need an evaluation method that holds up the same way for every candidate, every time, whoever is running the process.

The cost of getting it wrong compounds at scale as well. Each fraudulent hire is not only a salary and a ramp period lost. It is a security exposure, a team disruption, and in regulated environments a governance and audit question about how the decision was made. That is why a defensible, consistent evaluation method matters more than any individual detector. When you can point to what a candidate actually did, on a task relevant to the role, the decision is easier to explain and easier to stand behind.

Reframing fraud this way also connects to the wider truth that better evidence produces better hiring. The same job-relevant evidence that makes fraud expensive is the evidence that helps you identify the people most likely to succeed. You are not adding an anti-fraud step on top of hiring. You are hiring on the strongest evidence available, and resistance to fraud comes with it.

How Vervoe fits

Vervoe is an AI-powered skills intelligence platform, and this is the shift it is built for: moving hiring decisions off inferred ability, from resumes and self-reported profiles, and onto demonstrated ability. Instead of asking candidates to describe what they can do, Vervoe asks them to prove it through realistic, job-related tasks.

In practice, that means AI-powered screening, autograded skills assessments, and job simulations that show how a person performs the kind of work the role actually involves, graded consistently at scale so every candidate is evaluated against the same criteria. Because the evidence is performance on relevant work rather than a claim or a conversation, it is far harder to fake, and it stays useful long after the fraud tactics have changed. It also gives hiring teams something a detector never can: a clear, explainable record of what each candidate demonstrated, which supports more consistent and defensible progression decisions.

Skills assessments and job simulations will not, on their own, prevent every attempt at fraud, and no responsible vendor should claim otherwise. Sensible integrity controls and human judgment still belong in the process. What performance-based evaluation does is change the terms. It makes the most reliable way to pass your process the same as the ability to do the job.

Where to start

You do not need to rebuild your hiring process to act on this. A practical first move is to audit where your current decisions actually rest. For your highest-volume or highest-risk roles, ask how much of the progression decision depends on the resume and the interview, and how much depends on evidence of the candidate doing relevant work. If the honest answer leans heavily on the first two, that is your exposure.

From there, introduce a realistic, job-related task early in the process for the roles where a bad hire costs the most. The goal is not to catch fraudsters. It is to make demonstrated skill the thing that decides, so that catching fraud becomes a smaller and less urgent problem.

The defense that outlasts the next fake

Fraud tactics will keep improving, and detection tools will keep chasing them. That race is worth staying in at a basic level, but it is not where the hiring decision should be won. The organizations that handle candidate fraud well in 2026 will be the ones that stopped treating it as a detection problem and started treating it as a question of evidence: weighting their decisions toward performance that is genuinely hard to fake, because faking it means doing the work. That is a defense that does not expire the next time the fakes get better.

Explore how skills assessments and job simulations work, a lower-commitment starting point than a full demo.