Deepfake and proxy interviews: what employers can do.

A proxy interview is one where someone other than the applicant takes part. The stand-in might answer every question while the real applicant later does the job, or a helper might feed answers from off camera. A deepfake interview goes a step further: software alters the face or voice on the video call so that the person speaking appears to be someone else.

Both are forms of interview fraud, and both aim at the same gap: the interview is often the only time an employer sees a remote candidate before making an offer. In a Gartner survey of 3,000 job candidates, 6% admitted to participating in interview fraud, either posing as someone else or having someone else pose as them1.

Hiring managers report it more often than that figure might suggest, partly because one fraudulent candidate can apply to many companies. In Checkr's 2025 survey of 3,000 managers, 35% had confirmed that someone other than the listed applicant participated in a virtual interview2. In its 2026 survey, 67% had confirmed or strongly suspected an interview proxy standing in for a candidate, and 38% reported AI-generated or deepfake interview content among the fraud tactics they caught3.

These are surveys of what managers believe they have seen. They show the problem is widespread enough to plan for. They do not mean that a large share of the candidates you interview are fake, and most people on the other side of the camera are exactly who they say they are.

In June 2022 the FBI's Internet Crime Complaint Center warned of an increase in complaints about deepfakes and stolen personal information being used to apply for remote work, mostly in IT, programming, database and software roles. Some of those roles included access to customer personal data, financial data and corporate systems4. The complaints described voice spoofing during online interviews, lip movement that did not completely match the audio, and coughing or sneezing that did not line up with what was on screen4.

In January 2025 the FBI said North Korean IT workers had been observed using artificial intelligence and face-swapping technology during video job interviews to obscure their true identities5. CrowdStrike, a security vendor, reported that a North Korean group it tracks infiltrated more than 320 companies in twelve months, and that its IT workers reportedly use real-time deepfake technology to mask their identities in video interviews6. Our guide to North Korean IT worker schemes covers the wider operation.

Identity standards now treat this as a known threat. NIST's identity proofing guidelines list a deepfake video used to impersonate the person on a stolen driver's license as an example threat, and name presentation attack detection, which confirms the genuine presence of the applicant, among the mitigations7.

No single observation proves that an interview is fraudulent. These are the signs that tend to prompt a closer look:

• Audio and video out of step. Lips that do not match the words, or sounds such as a cough with no matching movement on screen4.
• A different person between rounds. The face, voice or manner of speaking changes noticeably between the phone screen, the first interview and later rounds.
• Relayed answers. A steady pause before every answer, eyes repeatedly moving to one spot off screen, or answers that sound read aloud and cannot be expanded when you ask a follow-up.
• Thin knowledge of their own history. The candidate cannot talk in specific terms about projects, colleagues or places on their own résumé. The FBI suggests soft questions about location and education background5.
• Avoiding video. Repeated refusal to turn on a camera, or pushing to move to text chat. US guidance on North Korean IT workers notes their preference for text-based chat over video calls8.

Each of these has an innocent explanation. A slow connection causes audio lag, nerves cause pauses, and some people need their camera off for disability, privacy or bandwidth reasons. An accent, a name or a country of origin is never a sign of fraud.

Interviewers are hired to assess skills and fit. Asking them to also act as fraud detectors puts them in a difficult position, and the results are uneven. In Checkr's 2025 survey, 62% of hiring professionals said job seekers are now better at faking their identities with AI than hiring teams are at detecting them2. In 2026, only 23% were extremely confident their process would catch a fraudulent candidate3.

Visual judgment has two failure modes. It misses good fakes, and it raises suspicion about genuine candidates with poor cameras, unstable internet or an unfamiliar way of speaking. Confronting a candidate mid-interview on a hunch can be humiliating for an honest person and simply alerts a dishonest one.

Timing matters too. Checkr found that only 29% of companies verify identity as soon as someone applies, while 49% wait until after at least one live interview3. By then a proxy may already have passed the stage where they were most likely to be noticed.

A more reliable approach takes the identity question out of the interview. Confirm who the candidate is through a separate step before they reach the interview, then confirm that the same person returns at each later stage. Interviewers can then focus on the job.

1. Verify identity before the first live interview. Ask shortlisted candidates to complete an identity verification step that checks a government ID against the person presenting it, with a liveness check to confirm a real person is present. Our guide to identity verification in hiring explains the options.
2. Re-check at each step that matters. A short check before the final interview and on the first day confirms that the person who verified is the person who turned up. The FBI recommends identity verification during interviewing, onboarding and throughout the employment of any remote worker5.
3. Keep the same interviewer where possible. Someone who met the candidate in an earlier round is best placed to notice a change.
4. Use structured, specific questions. Ask every candidate for the same role the same core questions, including ones that draw on details from their own résumé. A consistent script is fairer and makes unusual answers easier to compare.
5. Meet in person when it is practical. The FBI recommends completing as much of the hiring and onboarding process as possible in person5.
6. Watch the handover to employment. Be alert to requests to ship equipment to an address that is not on the person's identity documents8.

When something seems wrong

Finish the interview normally. Afterwards, write down what you observed and pass it to a named reviewer. The reviewer can ask the candidate to complete a verification step, or another one if they already have, and should make the decision with all the information in front of them. Avoid automatic rejection on the strength of an interviewer's impression.

If you believe you have encountered an organized scheme, the FBI asks companies to report it at ic3.gov, including any IP addresses, email addresses, phone numbers or names used4.

Checks that genuine candidates find intrusive or confusing will cost you good applicants. Tell candidates in advance what will be checked, when and why, and ask for clear consent before any face or ID capture. Explain what happens to their data and how long it is kept.

Offer an alternative route for anyone who cannot or prefers not to complete a camera-based check, such as sending their ID for review by a person, and make sure choosing it does not count against them. Apply the same process to every candidate at the same stage, regardless of name, accent or location. Consistency protects candidates from bias and protects your team from accusations of it.

Elsin verifies identity before the interview so the interview itself can stay an interview. Shortlisted candidates check their ID and face on their own phone; face images, video and ID numbers stay on the phone, and only the result, a face match score, timing evidence from the liveness check and a one-way code of the document number are sent. A quick re-check before the interview and on day one confirms it is the same person. It does not analyze the video call. Candidates who prefer not to scan their face can send their ID to a person instead, and nobody is rejected automatically.