TL;DR: Interview fraud is deception about who is taking an interview or who is producing the answers: a stand-in, an off-camera helper, AI-fed answers or a fabricated identity. In a Gartner 2Q25 survey of 3,000 candidates, 6% admitted to it, and the FBI and DOJ have documented organised schemes. No single check is conclusive. The defensible approach layers identity checks, interview design and human review, and treats a flag as a trigger for verification, not a verdict.
Updated 28 October 2026. Added the 13 October 2026 DOJ action; replaced the product section.
Key takeaways
- Gartner's 2Q25 survey of 3,000 candidates found 6% admitted to interview fraud (Gartner, 31 Jul 2025). Gartner's "1 in 4 candidate profiles fake by 2028" is a prediction, not a measured rate.
- The FBI has documented facilitators attending virtual interviews on behalf of North Korean IT workers, and warns that companies outsourcing IT work to third-party vendors are "removed from the direct hiring process" (FBI IC3, 23 Jul 2025).
- Human interviewers are poor detectors. In an interviewing.io test, in none of 32 analysed interviews did the interviewer suspect AI help, and Checkr's survey of 3,000 US hiring managers found only 19% "extremely confident" they would catch fraud.
- Defences work in layers (identity, interview design, behavioural signals, human review). Every signal has innocent explanations, so a flag should trigger a verification step, never an automatic rejection of the person.
- Monitoring candidates carries legal duties, including notice, consent and the EU ban on emotion inference. Build the process with counsel.
A face on a video call does not prove who is behind it, or who is writing the answers. This guide sets out what the public record shows, what the existing numbers measure, and which checks work and where each one goes wrong.
The question is not confined to human-led interviews. It follows screening wherever screening happens, including AI interview software, so the controls below apply in both settings.
What is interview fraud?
Definition. Interview fraud is deception about who is taking an interview or who is producing the answers. It takes five forms: a proxy or impersonator, an off-camera helper, AI-fed answers, fabricated credentials and a synthetic identity. It is distinct from preparation, which every candidate does.
The five forms:
- Proxy or impersonation. Someone else takes the interview in the candidate's place.
- Assisted interview. The candidate appears on camera while an off-camera helper supplies answers.
- AI-fed answers. Answers are generated or prompted by software during the interview.
- Credential fabrication. Employment or education history is invented or embellished.
- Synthetic identity. A constructed identity that does not correspond to a real applicant.
This page covers deception inside the interview and assessment. It does not cover fake-recruiter scams aimed at job seekers.
Proxy interview vs interview coaching
In a proxy interview, someone else answers. In coaching, someone prepares you and you answer. The line sits at who is speaking when the evaluation happens.
The target here is the fraud, not the candidate pool. In Gartner's survey below, 6% of candidates admitted to it, and some innocent behaviours resemble red flags, as the detection sections show.
How common is interview fraud?
Four sources give numbers, and each measures something different.
| Source | Date | Who was asked | Finding | Caveat |
|---|---|---|---|---|
| Gartner 2Q25 survey (confirmed via HR Dive, 8 Aug 2025) | 31 Jul 2025 | 3,000 candidates | 6% admitted to participating in interview fraud (HR Dive describes this as posing as someone else or having someone else pose as them) | Self-reported. |
| Gartner prediction (same release) | 31 Jul 2025 | n/a | 1 in 4 candidate profiles worldwide fake by 2028 | A prediction. "Fake profiles" is not "deepfake interviews." |
| Checkr, "Hiring Hoax" | 16 Sep 2025 | 3,000 US hiring managers | 31% interviewed someone later found to be using a false identity; 19% extremely confident they would catch fraud; 23% report losses over $50,000 | Vendor survey; methodology not detailed on the page |
| Pindrop | 18 Jun 2025 | Applicants to its own remote roles | 1 in 6 applicants showed signs of fraud | One employer's pool; sample for the 1-in-6 figure not stated. Pindrop reported this for its own openings; it is not an industry rate. |
The surveys measure self-admission, managers' recollection, a forecast and one employer's applicant pool, so the figures should not be compared or averaged. The methods side by side are in deepfake candidates and the documented record.
What do FBI and DOJ records show?
The table below lists the documented events in date order, with the primary source for each and what each one does not establish. It answers the question most readers arrive with: is this real, and how big is it?
| Date | What happened | Who says so | What it does not show |
|---|---|---|---|
| 28 Jun 2022 | FBI reported an increase in complaints of deepfakes and stolen personal data used to apply for remote jobs; named tells include lip movement out of step with audio and coughing or sneezing not matching the video | FBI IC3 PSA | A prevalence rate. Complaints are not a measure of all applicants. |
| 21 Apr 2025 | One researcher with no image-manipulation experience and a five-year-old computer built a synthetic interview identity in 70 minutes | Unit 42, Palo Alto Networks | That such identities are being used at scale. It is a demonstration of ease. |
| 18 Jun 2025 | Pindrop reported signs of fraud in 1 in 6 applicants to its own open roles | Pindrop | An industry rate. One company's pool; sample for the 1-in-6 figure not stated. |
| 30 Jun 2025 | DOJ announced coordinated nationwide actions across 16 states against North Korean remote IT worker schemes: two indictments, one arrest and one guilty plea, with seizures of 29 financial accounts, 21 websites and about 200 computers | DOJ | Charges are allegations apart from the one guilty plea. |
| 23 Jul 2025 | FBI stated that U.S.-based facilitators provide "attendance at virtual interviews and meetings on behalf of North Korean IT workers," and that sometimes one individual is employed to pass the initial interview while a different individual does the work | FBI IC3 PSA I-072325-4-PSA | How many interviews were affected |
| 31 Jul 2025 | Gartner survey: 6% of 3,000 candidates admitted interview fraud; prediction of 1 in 4 fake profiles by 2028 | Gartner | Anything beyond self-report; the 2028 figure is a forecast |
| 16 Sep 2025 | Checkr survey: 31% of 3,000 US hiring managers interviewed someone later found to use a false identity | Checkr | Verified incidents; this is managers' recollection in a vendor survey |
| Jun 2026 | Infosys postponed online assessments and in-person evaluations for 20,000+ trainee-role candidates after detecting impersonation. People Matters reported that the company said it put in place "additional controls and verification checks designed to protect the integrity of its recruitment assessments," and that The Times of India first reported the story | People Matters, 9 Jun 2026 | A video interview. It was an assessment event for trainee roles. |
| 13 Oct 2026 | DOJ announced coordinated actions against North Korean remote IT worker schemes affecting 70+ U.S. companies: 14 individuals charged, 2 arrests, about 187 devices seized, searches and other law enforcement action at over 40 residences in Texas known or suspected to be laptop farms | DOJ | Interview fraud or deepfakes. The release does not mention either. |
Three limits apply to the whole table. The DOJ entries describe charges and allegations; the defendants have not been convicted. No row is a measured industry rate, and the rows differ in kind: government warnings, vendor research, surveys and one company's exam disruption. The only entry that names a specific IT-services company concerns trainee assessments, not video interviews.
Why does proxy fraud show up in IT services and staffing chains?
The FBI's 2025 notice states that companies that outsource IT work to third-party vendors "can face additional vulnerabilities since these companies are removed from the direct hiring process." It also describes the pattern directly: an individual is employed to pass the initial interview, but a different individual completes the on-the-job work (FBI IC3, 23 Jul 2025).
The U.S. and North Korea context here is evidence of a structural pattern only. It implies nothing about any named IT-services company. The full treatment is in proxy fraud in IT-services hiring.
How do you spot a fake candidate in a video interview?
Treat every sign as a trigger for a verification step. Three probes come from government and security sources:
- Camera on, background unobscured. The FBI recommends mandating video and requesting an unobscured background (FBI IC3, 23 Jul 2025).
- A hand-over-face probe. The FBI suggests asking for a hand wave in front of the face, which may cause AI-generated video to malfunction. Unit 42 calls hand over face the most effective probe.
- Image capture for comparison. Capture images at each round so later meetings can be checked against them (FBI, same notice).
These are recommended practices, not validated detection methods, and their efficacy has not been independently confirmed. Ask any vendor selling face or liveness checks how it tests against current real-time video tools.
Honest candidates trip these probes too. Disability, neurodivergence, poor connectivity and caregiving at home can all produce a camera that is off, a gaze that drifts or a delay. The full flag-to-verification-step checklist is in how to spot fake candidates.
What about candidates using ChatGPT or real-time AI help?
Preparing with AI beforehand is ordinary. Real-time help during an evaluation of present ability is the contested line. Employers can ban it, allow it with disclosure, or redesign the interview so that it matters less. The policy framing is in candidates using AI in interviews.
Interviewers are poor at noticing it. In the interviewing.io experiment (37 interviews, 32 analysed), no interviewer reported suspecting cheating in the 32 analysed interviews. Pass rates were 73% on verbatim LeetCode questions and 25% on custom questions, against 53% for the control group. The experiment is undated (about 2023) and small, and should be read that way.
The tools themselves, and what their sellers claim, are covered in AI-coached cheating.
The four layers of defence, and what each one gets wrong
No single control is conclusive. Each of the four layers fails in a characteristic way.
| Layer | What it checks | Common innocent cause of a false flag | Cost or limit |
|---|---|---|---|
| Identity | Document and photo comparison across rounds | Different device, different lighting, name changes | Notice and consent duties vary by jurisdiction |
| Interview design | Unscripted follow-ups, a different question per round, an in-person round for finalists | Not applicable; the risk is burden | Candidate burden |
| Behavioural signals | Presence, gaze, phone | Neurodivergence, accessibility needs, shared rooms | Vendor-reported; ask each vendor for its false-positive rate and how it was measured |
| Process and records | Who verified what, at which stage; retained evidence; an appeal path | Records can carry an earlier mistaken flag forward | Needs an owner and a retention rule |
Behavioural signals are the layer software can log, and the layer where vendor claims need the most scrutiny. How to evaluate vendors is covered in interview cheating detection software.
The second table maps each form of fraud to the control the cited guidance points to, and to what trips that control. The mapping is editorial. No source rates these controls against each other.
| Form of fraud | Identity | Interview design | Behavioural signals | Process and records |
|---|---|---|---|---|
| Proxy or impersonation | Image comparison across rounds (FBI); false flag: new device or lighting | Unscripted follow-ups on the candidate's own answers | Face absent, multiple faces; false flag: shared room | Record which round each check covered |
| Assisted (off-camera helper) | Limited; the right person is on screen | Custom questions and follow-ups | Gaze and phone events; false flag: notes, accessibility needs | Record who reviewed the flag |
| AI-fed answers | Limited | Custom questions cut the cheater pass rate from 73% to 25% (interviewing.io) | Reading gaze may register; false flag: notes, neurodivergence | Record the policy the candidate was shown |
| Credential fabrication | Verify employment and education directly with institutions (FBI); false flag: mergers, contractors | Probing follow-ups on claimed experience | Not relevant | Retain the verification result |
| Synthetic identity | Cross-reference documents and photos with social profiles (FBI); newly created profiles (Pindrop); false flag: a privacy-minded candidate | In-person round for finalists | Not relevant | Escalate to an extra ID check with HR |
The applicant-tracking side offers one data point that identity checks are treated as table stakes. Greenhouse's November 2025 report talks about introducing "good friction" like checking identity (Greenhouse). Greenhouse also markets Real Talent on its product pages; ask any ATS vendor which identity checks it runs and what they cover. Behavioural signals sit alongside ATS-side identity checks; they do not replace them.
In our reading, AI-text detectors are not a safe control for interview answers: spoken answers are not written text, and a detector score is one more flag with innocent explanations, not proof.
What does the law say about monitoring candidates?
Notice and consent rules are jurisdiction-specific. In the EU, Article 5(1)(f) of the AI Act bans AI used to infer emotions of a person in the workplace, in force from 2 Feb 2025, with exceptions for medical or safety use (AI Act, Art. 5). Any "behavioural monitoring" must therefore avoid emotion inference. The EU AI Act recruitment rules go further into what the Act covers in hiring.
This page has not verified GDPR Article 9 on biometrics, Illinois's BIPA or state video-interview statutes, and it should not be read as covering them. In the United States, tools that influence decisions may also fall under NYC Local Law 144, and recording and disclosure obligations are covered in interview recording consent rules. Check the Illinois video-interview rules and any other state rules before running a process.
Is it legal to monitor candidates? It depends on jurisdiction and method. Disclose monitoring in the invitation, and keep a human in the review of any flag.
Where ZeaHire fits
ZeaHire reports behavioural proctoring signals as evidence for a human reviewer. In autonomous AI interviews, the proctoring report counts look-aways, head turns, suspected reading, phone use, face absent, multiple faces, liveness issues, screen overlap and mouse clicks. An Evidence tab lets a reviewer inspect them, and penalty settings are configurable. The candidate record carries manual Approve and Reject controls for the reviewer.
These signals are evidence for review. They are not proof of fraud, and they do not establish who the candidate is. To see how they are presented, see an evidence record in a demo.
FAQ
What is interview fraud?
Interview fraud is deception about who is taking an interview or who is producing the answers. It includes proxy stand-ins, off-camera helpers, AI-fed answers, fabricated credentials and synthetic identities. This page covers deception in the interview and assessment, not fake-recruiter scams aimed at job seekers.
What is a proxy interview, and how is it different from interview coaching?
A proxy interview is one where someone other than the candidate answers. Coaching is preparation: someone helps the candidate get ready, and the candidate answers. This page does not describe how proxy services operate.
How common is interview fraud?
Gartner's 2Q25 survey of 3,000 candidates found 6% self-reported taking part in interview fraud. Checkr's survey of 3,000 US hiring managers found 31% had interviewed someone later found to be using a false identity. The figures differ because one is candidates' self-admission and the other is managers' recollection in a vendor survey.
How can I tell if a candidate is a fake?
No single sign proves it, and most have innocent explanations. Treat each sign as a trigger for a verification step, such as a second call, an unscripted follow-up or a document check, and never as grounds for rejection on its own.
Can candidates use ChatGPT in an interview?
It depends on the employer's policy. In our reading, employers take one of three positions: prohibit live AI help, allow it with disclosure, or redesign the interview around it. Anthropic's candidate guidance, for example, tells applicants that live interviews are "all you", with no AI assistance unless the company indicates otherwise (Anthropic, updated 10 Jul 2025).
Is it legal to monitor candidates during interviews?
It depends on jurisdiction and method. Disclose monitoring in the invitation, avoid emotion inference in the EU, and have counsel review the process. The state-by-state AI hiring rules are a starting point for the US.
What should you do when you suspect a proxy?
Pause the process and verify identity through a second channel. Document what was observed, and involve HR and legal. Do not accuse the candidate in the interview.
What is AI interview proctoring, and how does it work?
Proctoring means monitoring a remote session for signs that the person or conditions differ from what was agreed. In interviews, it generally means software that logs events during the session, such as whether a face is in frame, for a person to review.
Next step
See what a ZeaHire evidence record contains: transcript, per-question scores and proctoring counters, laid out for a human reviewer. See an evidence record in a demo.
Comments
No comments yet — start the conversation.