Deepfake job candidates: what the FBI warns and how to check

Deepfake job candidates use AI-generated or face-swapped video, and sometimes altered voices, to pass remote interviews as someone else. The FBI has warned about it since 2022 and said in January 2025 that North Korean IT workers "have been observed using artificial intelligence and face-swapping technology during video job interviews." Recruiters can run quick live checks, such as asking the candidate to wave a hand in front of their face or turn their head, and should pair them with an identity check and a record they can compare against later. None of these checks is proof on its own.

What has the FBI warned about deepfake candidates?

Three FBI public service announcements cover this, and each adds something.

The 2022 advisory also noted that some roles targeted this way "include access to customer PII, financial data, corporate IT databases and/or proprietary information." That is the real concern. A fake candidate for a role with no access to anything valuable is a nuisance. A fake candidate with database access is a security incident.

Hiring managers are worried about it too. Checkr's September 2025 survey of 3,000 US managers lists deepfake video interviews among the tactics causing the most worry.

What does a deepfake look like on a live call?

Don't expect obvious glitches. What tends to give it away is motion and timing, the things the software has to compute live.

• Mouth movement that drifts out of sync with the voice, which is the pattern the FBI described in 2022
• Coughs, laughs or sneezes that you hear but don't see
• Blurring or warping at the edges of the face, hairline, glasses or ears, especially when the head turns
• Lighting on the face that doesn't match the lighting in the room behind it
• A face that looks oddly still or smooth while the body moves normally
• A background that is blurred, replaced, or kept out of view, which the FBI's July 2025 advisory addresses by recommending you "request that their backgrounds be unobscured"

A poor webcam and a weak connection cause some of the same artifacts. A single glitch means little. Several at once, especially when the candidate also resists the checks below, is worth escalating.

What live checks can recruiters run?

These take a couple of minutes and fit into a normal video screen. Tell candidates at the start that you do a short identity check with everyone, so nobody feels singled out.

• Ask the candidate to wave a hand slowly in front of their face. The FBI suggests this because it "may prompt a malfunction in AI generated video."
• Ask them to turn their head to show each profile, then look down at the desk and back up.
• Ask them to hold their government ID next to their face, then compare the photo with the person and with photos on their resume, portfolio or professional profile.
• Ask them to point the camera out a window, then ask about their claimed location. The FBI's July 2025 advisory recommends asking "about their claimed current location and the location listed on their identification documents."
• Ask "soft" questions about location or education history, which the FBI's January 2025 advisory recommends. You are checking that the answers match the ID and resume. Where someone is from, or where they studied, is not a red flag.
• Capture images or keep the recording (with notice to the candidate) so you can compare the person at the client interview or on day one. The FBI notes that "sometimes an individual is employed to pass the initial interview, but the on-the-job work is completed by a different individual."
• For final rounds on sensitive remote roles, do part of the process in person if you can. The FBI's January 2025 advisory recommends completing "as much of the hiring and onboarding process as possible in person."

Before you record or keep images, check the rules where your candidates are. Our guide to AI hiring laws covers notice requirements for AI video interviews in some states. That guide, like this article, is general information, not legal advice.

Where do tools help, and where don't they?

Face-swap software keeps improving, so be specific about what a tool actually does.

Tools help in a few ways. They can tie the interview to an identity document at the start, so someone using a face swap also has to get through a government ID check and a live face match, which raises the effort. They can add views that are harder to fake at the same time. Face-swapping one webcam feed is one problem; faking a side angle from a separate phone as well is a harder one. And they keep a timestamped record with screenshots and audio, which gives you something concrete to compare when a different person shows up later.

Tools don't settle the question on their own. The hand-wave test is phrased by the FBI as something that "may" cause a malfunction, so a clean wave proves nothing. No automated check catches every fake ID or every altered video, and any vendor claiming a catch rate should be asked how it was measured. Voice cloning is also a factor. The FBI's 2022 advisory mentions "voice spoofing, or potentially voice deepfakes," and an audio-only screen gives you fewer signals than one with video and ID.

Human judgment is also the fairness check. Never treat an accent, a name or a candidate's apparent ethnicity as a deepfake signal; doing so is national origin discrimination as the EEOC defines it. People with facial differences, poor lighting, older laptops or slow connections can trip the same "glitch" signs as a deepfake. A person should review the evidence and ask for re-verification before anything is decided.

How Simplehire handles deepfake risk in screening

Simplehire runs AI voice screening interviews for recruiters, with a screening report you review before submitting a candidate. We don't run AI analysis of the interview audio or camera feeds to judge whether it is synthetic, and we don't claim to detect deepfakes. What we do is make impersonation harder and leave a clear record.

On Enterprise, you can set the ID check to Off, Optional or Required. When it is required, the candidate must pass a government ID check and a live face match before the interview begins. You can also require a paired second device. The candidate scans a code with their own phone, with no app and no account, and the phone gives a side-camera view of the candidate, their laptop and the surrounding desk. Lighting, framing and angle are checked before the interview starts, and candidates are told before monitoring begins.

Every tier runs in enforced full-screen mode and records tab switches, focus loss and copy and paste attempts. On supported Windows interviews, Simplehire Proctor has candidates close risky applications such as remote-desktop tools before starting. Integrity events are timestamped, interview audio is recorded, and alerts come with screenshots and a severity level. A person reviews every finding. Simplehire raises alerts; the recruiter decides.

Screens are $2 (Basic), $3 (Adaptive) or $5 (Advanced), with 25 free Basic screens every month and no credit card. See the proctoring details or start free.

Deepfakes are one form of impersonation. For the broader stand-in problem, see how to spot a proxy interview. If you place remote IT contractors, our North Korean IT worker checklist covers the scheme the FBI ties most closely to face-swapped interviews.