Candidate Fraud in Hiring: Fake Applicants, Deepfakes and AI-Written Answers

Candidate fraud detection means checking, at every stage of hiring, that the applicant is the person they say they are and has done the work they say they have. Most fraud falls into five patterns: fabricated profiles and résumés, proxy interviews, deepfaked video, AI-written answers read aloud, and a different person turning up at a later stage.

No single tool catches all five. What works is a small check at each stage, an interview that asks for detail a fraudster doesn't have, and a record you can compare from one round to the next. This guide covers each kind of fraud, the numbers behind the worry, and what to check from application to the first day.

How common is candidate fraud?

Nobody has a clean count, because fraud that succeeds mostly goes unnoticed. The direction is clear, though. Gartner forecasts that by 2028, one in four candidate profiles worldwide will be fake. That is a forecast, not a measurement. In a Gartner survey of 3,000 job candidates in the second quarter of 2025, reported by HR Dive, 6% admitted to interview fraud: posing as someone else, or having someone else pose as them. That is what people admit to in a survey.

Hiring teams say they see it. In Greenhouse's November 2025 survey, a vendor survey that included 665 US recruiters and hiring managers, 91% of US recruiters said they had spotted candidate deception, and 65% of US hiring managers said they had caught applicants using AI deceptively.

The risk isn't spread evenly. In 2022 the FBI's Internet Crime Complaint Center warned of more complaints about deepfakes and stolen personal information being used to apply for remote jobs in IT, programming, databases and software, some with access to customer data, financial data and company systems. A remote role with access to money, code or customer records deserves every check in this guide. A role you interview for in person needs fewer.

What counts as candidate fraud?

Candidate fraud is misrepresenting who you are, or what you have done, to get hired. Polish isn't fraud. A candidate who uses AI to tighten a résumé, practices answers with a chatbot, or blurs the kitchen behind them hasn't lied about anything. In the same Greenhouse survey, 36% of US job seekers said they had used AI to alter their appearance, voice or background in video interviews, and most of them said they did it to look more professional. Treat every filter as a fraud signal and you will flag a lot of honest people.

The line is identity and substance. Three questions cover it:

• Is this the person who applied?
• Did they do the work the résumé describes?
• Are the answers in the interview their own?

The five kinds of candidate fraud

Fabricated profiles and résumés

Some fake applicants are invented from scratch: a name, a profile, a résumé tailored to your job description, sometimes with a stolen identity underneath. Others are real people with an invented history, such as employers that don't exist, titles they never held, or project descriptions lifted from someone else. Generative AI makes both cheap to produce in volume.

A newer variant is aimed at AI screeners rather than people: hidden text in a résumé, often in white type, telling a screening model to rank the applicant highly. Greenhouse's survey lists these prompt injections among the deceptive uses of AI hiring managers had caught (22%).

What to check:

• The résumé against the candidate's public profile: employers, titles and dates.
• That the employers exist and the dates fit. A company founded in 2021 can't have employed someone in 2018.
• Contact details shared across applicants. The FBI's January 2025 warning about North Korean IT workers using false identities noted the same phone numbers, particularly VoIP numbers, and email addresses turning up on multiple résumés. It recommends cross-checking your hiring systems for applicants with matching résumé content or contact details.
• Hidden text. Paste the résumé into a plain-text editor, or select all in the PDF, and see what appears.

Proxy interviewing

Someone other than the applicant takes the interview, or sits out of view feeding answers. This is the fraud behind Gartner's 6% figure. It shows up as a mismatch: the voice in the technical round isn't the voice from the screen, or the person on the first day can't do what the person in the interview could.

What to check: whether the candidate can talk about what was said in an earlier round, whether voice and manner match across rounds, and whether the answers hold up when you ask about the résumé in detail.

Deepfakes on video

A real-time deepfake swaps a face or a voice during the call. The FBI's 2022 warning described what complainants noticed: lip movements that didn't fully match the audio, and coughs or sneezes that didn't line up with what was on camera. Its 2025 warning described applicants using AI and face-swapping technology during video interviews to hide their identity.

What to check: audio against lip movement, and what happens when the person moves. Experts quoted by Fortune in 2022 suggested asking the person to turn to a full profile, which removes about half the facial reference points the software tracks and often makes it warp, or to pass a hand or an object across their face, which some deepfakes fail to render. The same experts warned that deepfakes keep improving. A failed test is worth a closer look; a passed one proves little.

AI-written answers read aloud

The candidate hears the question, an AI tool on another screen drafts an answer, and the candidate reads it out. Of the examples Greenhouse listed, reading from AI-generated scripts was the most common (32%). The usual signs are a pause before every answer followed by a fluent, well-organized answer that is oddly general; vocabulary that doesn't match how the person talks otherwise; answers that restate the question; and, on video, eyes tracking across a screen.

None of those signs proves anything. Nervous candidates pause. People interviewing in a second language plan their sentences before they speak. Some candidates keep notes. The real test of AI interview cheating is the follow-up question, covered below.

Identity mismatches between stages

The person who passed the screen isn't the one in the final round, or the one who interviewed isn't the one who starts. Sometimes a proxy interviewed and someone else does the job. Sometimes a fake identity holds up until the paperwork. The FBI's 2022 warning noted that pre-employment background checks found some applicants were using personal information that belonged to someone else, and its 2025 warning told employers to watch for changes of address or payment platform during onboarding.

What to check: the same person, end to end. Keep the record of each round so you can compare them, and make sure the identity check at offer covers the person you actually interviewed.

How to detect and prevent candidate fraud at each stage

Fraud prevention works best as small checks spread across the process, not one big gate. This table is a starting point; scale it to the risk of the role.

Two habits make the table work. First, apply the same checks to everyone for a role, not just to candidates who seem off; consistency is fairer to candidates and harder to game. Second, move the expensive checks, like in-person meetings and identity verification, to the late stages, where only a few candidates remain.

Many teams are doing exactly that: in the same Greenhouse survey, 39% of US hiring managers said they were holding more in-person interviews to verify authenticity. Gartner's advice, as HR Dive reported it, is a layered approach: clear expectations about acceptable AI use, assessments such as in-person interviews, and, after hiring, tighter background checks, identity verification and anomaly alerts.

Why follow-up questions, résumé consistency and a recording make fraud harder

Most fraud is cheap to start and expensive to keep up. Three parts of a good interview raise that cost.

Live follow-up questions

A script is written for the question that was asked. A follow-up asks for what the script doesn't have. Someone who did the work goes one level deeper without effort; someone reading or relaying answers has to wait for the next draft, and the drafts tend to stay general. Ask at least two follow-ups on any claim that matters for the role:

• "What was the number before, and what was it after?"
• "Which tool did you use, and what did you set up yourself?"
• "Who else worked on it, and what was their part?"
• "What went wrong, and what did you change?"
• "Walk me through the last time you did that, step by step."

These are also the questions that make a structured interview work: the same core questions for every candidate, with follow-ups that probe for evidence.

Consistency between the résumé and the answers

Ask about what the résumé claims, and let the candidate describe it in their own words. Real work comes with small, uneven detail: the team was four people, not five; the migration slipped a month; the tool was replaced halfway through. A fabricated history tends to come apart at exactly those points: dates, team sizes, tools and who did what. A mismatch is a question to ask, not a verdict. People round up titles and misremember dates.

A recording a person can review

A recording turns a hunch into something you can check. A second reviewer can listen. The voice from the first round can be compared with the third. And candidates who know an interview is recorded and reviewed by a person have less reason to try. Tell candidates up front that the interview is recorded and who will review it; the candidate experience guide covers how to say it.

An AI interview can make all of this easier or harder. One that scores résumé keywords, or a recorded monologue with no follow-up, gives a fraudster a fixed target. A structured voice interview with follow-ups gives a recruiter more to check than a résumé does. DIO, the AI interviewer Dio makes, asks follow-up questions when an answer is thin or vague. Each competency score quotes the candidate's exact words, and the quote is checked against the transcript. The team gets the recording and the transcript to review. None of that is a verdict on fraud. It is evidence a person can hold up against the résumé and against the next round. How AI interview scoring works explains why quoted evidence matters.

How to avoid flagging honest candidates

Every false alarm lands on a real person, usually one who did nothing wrong. A few rules keep fraud checks fair:

• Treat a signal as a reason to ask, not to reject. A follow-up question costs a minute. A wrong accusation costs a candidate and your reputation.
• Don't count accent, a slow start, background blur or notes as fraud on their own. Each has plenty of innocent explanations.
• Say what AI use is acceptable. Gartner recommends setting clear expectations about acceptable AI use. A plain version: using AI to prepare is fine; using it to answer during the interview is not.
• Tell candidates what you check. Gartner also suggests communicating about your fraud detection efforts. It can put off some fraud, and honest candidates know where they stand.
• Let a person decide, with the evidence in front of them. Write down what you saw and why it mattered, so the next reviewer can check your reasoning.

Candidate fraud detection tools: what to expect

Candidate fraud detection tools fall into a few groups: identity verification (a photo ID matched to a live selfie), deepfake and liveness checks during video calls, employment and education verification, duplicate-applicant checks in your applicant tracking system, and monitoring software that watches the screen or browser during an assessment. Each can add a signal. None replaces the basics above.

Before you buy one, ask the vendor:

1. What does it check, at which stage, and what does the recruiter see?
2. How often does it flag an honest candidate, and how was that measured?
3. Can a person see why something was flagged?
4. How are candidates told about the check, and how can they respond to a flag?
5. What happens to the data it collects, and when is it deleted?

If a vendor can't answer the second question with a method, treat its accuracy claims with care. For the formats themselves and how much each can verify, see AI screening interviews compared and what an AI interviewer does.

The short version: verify early, probe live, compare across rounds, and confirm identity before the offer. Fraud gets through when one stage assumes another stage already checked.