New — The State of AI Hiring 2026. Every number sourced→
Data Security

How to Spot Fake Candidates in Video Interviews: A Red-Flag Checklist Mapped to Verification Steps

How to spot fake candidates in video interviews: red flags, innocent explanations and the verification step for each, sourced to FBI and Gartner guidance.

How to spot fake candidates in video interviews, with a card pairing red flags with the step that settles each

TL;DR: How to spot fake candidates in video interviews: a fake candidate is someone who is not who they claim to be: a proxy, an impersonator or a synthetic identity. No single sign proves it, and most signs have innocent explanations, so treat each one as a trigger for a specific verification step (rejoin on a second call, an unscripted follow-up, a document check, image comparison at each round), never as grounds for rejection. The table below pairs each sign with its innocent explanation and its next step.

Updated 28 October 2026; reflects FBI IC3 guidance of 23 Jul 2025.

Key takeaways

  • The FBI's 23 Jul 2025 notice lists concrete video checks: camera on with an unobscured background, pointing the camera out a window to discuss location, a hand wave in front of the face, and image capture for comparison across later meetings (FBI IC3).
  • Gut feel is not a control. In the interviewing.io experiment, in none of 32 analysed interviews did the interviewer report suspecting a ChatGPT-assisted candidate, and only 19% of 3,000 US hiring managers in Checkr's survey were "extremely confident" they could catch fraud.
  • Every red flag has a plausible innocent cause (lag, hardware, nerves, disability, neurodivergence, caregiving, poor bandwidth), so the right response is an extra verification step.
  • A 2Q25 Gartner survey of 3,000 job candidates found 6% admitted to participating in interview fraud (Gartner, 31 Jul 2025). The self-reported rate is low, the cost of a false accusation is real, and most signs have an innocent cause, so a flag should lead to a verification step, not a verdict.
  • Designing the interview so a stand-in struggles (unscripted follow-ups on the candidate's own answers) beats trying to read faces.

You are partway through a video interview and something is off. The lips trail the audio, the answers arrive after a pause, the face softens when the candidate turns. Each of those could be a fake candidate or a weak connection, and the sign alone cannot say which. This page pairs each sign with its innocent explanation and the step that settles the question.

The scale and the documented record of proxy interview fraud are covered on the hub page. This page covers the other question: how to spot it, and what to do next. How it plays out when IT services and staffing firms hire at volume is covered in proxy fraud in IT services hiring.

What counts as a "fake candidate"?

A fake candidate is someone who is not who they claim to be. That covers a proxy or impersonator who takes the interview in someone else's place, a person using a fabricated or stolen identity, and a synthetic identity with no real applicant behind it. Embellished credentials on their own belong to background checks, not this page.

For scale, two lines. According to Gartner (31 Jul 2025), a 2Q25 survey of 3,000 job candidates found 6% admitted to participating in interview fraud, and Gartner predicts that "by 2028, one in four candidate profiles worldwide will be fake" (Gartner press release). The second figure is a prediction, not a measured rate. In Checkr's 2025 survey of 3,000 US hiring managers, 31% said they had interviewed someone later found to be using a false identity (Checkr).

Which red flags do government and security sources actually name?

Four sources name specific signs and probes. None of them claims that a sign proves fraud.

  • FBI IC3, 28 Jun 2022. Lip movement and actions that do not fully coordinate with the audio, and coughing or sneezing that does not line up with what is on screen (FBI IC3).
  • FBI IC3, 23 Jul 2025. Scrutinise identity documents for misspellings and cross-reference photos and contact details with social profiles. Verify prior employment and education directly with the institutions. Require in-person meetings where possible. On video, mandate a camera and an unobscured background, have the person point the camera out a window and discuss their location, and ask them to wave a hand in front of the face, which may cause AI-generated video to malfunction. Capture images for comparison with later meetings (FBI IC3).
  • Unit 42 (Palo Alto Networks), 21 Apr 2025. A hand over the face, which it calls most effective, plus rapid head movement, exaggerated expressions, a sudden lighting change and profile turns (Unit 42). These are recommended probes from security-vendor research. Independent validation of their effectiveness has not been confirmed; treat any claim about how fast real-time tools are improving as unverified.
  • Pindrop, 18 Jun 2025. Signals it uses on its own applicants before the interview: resumes listing shell companies, newly created LinkedIn profiles with few connections, suspicious IP addresses and forged documents (Pindrop). This is a single vendor's own applicant pool with the sample size unstated, so read these as signals, not rates.

The sourced record behind these probes is in what the record shows on deepfake candidates. No single probe is settled.

Why doesn't gut feel catch it?

Because interviewers are not good at it, and the evidence says so. In the interviewing.io experiment, 37 interviews were run and 32 analysed, and no interviewer reported suspecting that the candidate was getting help from ChatGPT. The cheating candidates passed 73% of the time on verbatim LeetCode questions and 25% on custom questions, against 53% for the control group (interviewing.io). The experiment is undated (about 2023) and small, so it shows a pattern, not a rate.

Hiring managers report the same gap. In Checkr's "Hiring Hoax" survey of 3,000 US hiring managers (16 Sep 2025), 31% said they had interviewed someone later found to be using a false identity, and only 19% were extremely confident they could catch fraud (Checkr). It is a vendor survey, and its methodology is not detailed on the page.

The same problem applies to real-time AI help, which is mostly invisible to an interviewer. It is covered under AI-coached answers.

The checklist: red flag, innocent explanation, verification step

Escalate on patterns across rounds, not on a single flag. Each row names the sign, what it might really be, the step that settles it, who does it, what not to do, and the source.

#Red flagWhat it might really beVerification stepWhoDo notSource
1Lip movement out of step with audioLag, a bad connectionAsk the candidate to rejoin; compare on a second callInterviewerAccuse, or infer deception from the glitchFBI 2022
2Face distorts on a hand pass, fast head turn or lighting changeWebcam compressionRepeat the probe after a camera reset; if it persists, move to a verified in-person or ID step (editorial guidance, not from either source)InterviewerReject on one failed probeFBI 2025 (hand wave only), Unit 42 (hand, head, lighting probes)
3Camera "broken," audio-only requests, repeated platform switchesGenuine hardware or policy limitsOffer an alternative verified slot; watch for a pattern across roundsRecruiterTreat one request as proof; ignore disability or bandwidthPractitioner observation; no published data behind it
4Fixed-delay answers, reading gazeNotes, nerves, accessibility needsAsk an unscripted follow-up on the candidate's own earlier answer; change the question orderInterviewerInfer emotion from gaze (EU AI Act Art. 5)interviewing.io (custom questions)
5Cannot describe the claimed locationRecently movedAsk the candidate to point the camera out a window and give local specificsInterviewerTreat unfamiliarity as disqualifyingFBI 2025
6Face or voice differs from round to roundPoor lighting, a different deviceCapture images at each round and compare; match again at onboardingRecruiter or HRDecide on one comparison; skip telling candidates in the invitationFBI 2025 (image capture for comparison only)
7The listed employer cannot confirm the roleMergers, contractorsVerify employment directly with the institutionRecruiter or background vendorTreat a failed lookup as a lieFBI 2025
8New profile, few connections, IP tied to known proxy infrastructureA privacy-minded candidateEscalate to an extra ID checkRecruiterReject on profile age alonePindrop (signals from its own pool)

What should you verify, and when?

Verification works best in three stages.

Three verification stages, before, during and after the interview, each listing FBI-sourced checks, with a note that a flag is a trigger, not a verdict

  • Before the interview. Check documents and profiles for the signals in rows 7 and 8. Verify employers and education directly with the institution, and look at how new the profile is.
  • During the interview. Run the FBI's video checks: camera on, background unobscured, camera out the window for a location question. Add an unscripted follow-up on something the candidate said earlier.
  • After each round and at onboarding. Compare captured images across rounds and at onboarding. The FBI ties image capture to the pattern in which one person passes the interview and a different person does the work (FBI IC3).

Capturing and storing images

Disclosure and retention rules for images are jurisdiction-specific and need counsel. Tell candidates in the invitation that identity checks are part of the process, and retain only what the check needs. This page does not assert coverage under any specific statute; the Illinois AI Video Interview Act, BIPA and GDPR Article 9 were not researched for this page. For disclosure, see consent and retention for ID images.

A flag is a trigger, not a verdict

Each flag has innocent explanations: disability, neurodivergence, caregiving, poor connectivity, and accents or backgrounds that an interviewer is not used to. The instruction is the same each time. A flag leads to an extra verification step, and a flag alone never leads to rejection.

Do not use lie or emotion detection. Article 5(1)(f) of the EU AI Act bars AI from inferring emotions in the workplace from 2 Feb 2025 (AI Act, Art. 5). Avoid certainty language as well. One published vendor guide lists "immediate disqualifiers." That framing treats a flag as a verdict, and it is the framing this page rejects. Nothing here is a disqualifier list.

Design the interview so a stand-in struggles

The strongest control is not a better look at the screen. It is a better interview.

  • Ask unscripted follow-ups on the candidate's own answer. Prepare unscripted follow-up questions by theme, then pick them live.
  • Change question order and content per round. In the interviewing.io experiment, custom questions cut the cheater pass rate from 73% to 25%.
  • Include one live reasoning segment, where the candidate works a problem aloud.
  • Add an in-person round for finalists where it is proportionate to the role.

Anthropic's engineering team went through the same exercise for its own take-home. Tristan Hume reported that the test was redesigned repeatedly as models improved, and that tasks "out of distribution" from what models had seen, along with building one's own debugging tools, separated skill from delegation (Anthropic Engineering, 21 Jan 2026). Whether to ban, allow or redesign around AI help is a separate decision, covered in AI-use policy options for interviews.

Where software helps, and where it does not

Software can log presence, count events and keep records. It cannot settle identity by itself, and government guidance leans on human steps: documents, direct employer checks (FBI IC3). Ask any detection vendor whether it publishes false-positive rates for its signals.

In autonomous AI-led screening interviews, ZeaHire's proctoring report counts look-aways, head turns, suspected reading, phone use, face absent, multiple faces, liveness issues and screen overlap, and puts them in an Evidence tab for a human to review. Against the checklist, these counts touch only reading gaze (row 4). They do not detect face distortion or compare faces between rounds. They are signals for review. They do not establish who the candidate is.

FAQ

How can you tell if a candidate is using a fake background or filter?

Ask the candidate to rejoin or reposition the camera, and compare against a second call. A background effect or filter alone does not establish that the person is not who they claim to be.

What are the signs of a deepfake in a video interview?

The FBI names lip movement that does not match the audio and coughing or sneezing that does not line up with the picture. Unit 42 recommends a hand over the face, rapid head movement and sudden lighting changes as probes. Tools improve, and the absence of a tell proves nothing.

Is it normal for a candidate to keep their camera off?

Sometimes. Hardware problems, company policy, disability and poor bandwidth are all possible causes. Offer a verified alternative slot, and look for a pattern across rounds before drawing any conclusion.

How do I verify someone's identity before an interview?

Check identity documents for misspellings and cross-reference photos and contact details with social profiles. Verify prior employment and education directly with the institutions (FBI IC3). Consent and retention rules are jurisdiction-specific.

What do I do if I suspect a fake candidate?

Pause the process, verify through a second channel, document what you observed, and involve HR and legal. Do not make an accusation in the interview.

Can I ask a candidate to show ID on camera?

It depends on jurisdiction and on how long the image is kept. Disclose the step in the invitation, retain only what the check needs, and have counsel review the process.

How common are fake candidates?

Gartner's 2Q25 survey of 3,000 candidates found 6% self-reported interview fraud (Gartner). Checkr's survey of 3,000 US hiring managers found 31% had interviewed someone later found to be using a false identity. It is a vendor survey, so read it as a signal of how often managers say they have met the problem, not as a measured rate. The two figures use different methods and should not be compared directly.

Do fake candidates only target remote tech roles?

FBI warnings centre on remote IT roles (FBI IC3). The sources here do not establish how widely the pattern spreads across other roles, so treat the claim with care in either direction.

Next step

A ZeaHire evidence record holds the transcript, per-question scores and proctoring counters for a reviewer to read. See what a ZeaHire evidence record looks like.

Interview fraudHiring integrity
Was this useful?

Comments

No comments yet — start the conversation.

Leave a comment

Your email is never published. Comments appear once a moderator approves them.

Replying to
Plain text only. Be kind and stay on topic.

By commenting you agree to our privacy policy.