TL;DR: The documented record on deepfake candidates is thin and specific. The FBI warned in 2022 about deepfakes in remote-job applications, and a 2025 FBI advisory describes facilitators attending virtual interviews for North Korean IT workers. DOJ's October 2026 action against such schemes charges 14 individuals and does not mention deepfakes. Gartner's prediction that "one in four candidate profiles worldwide will be fake" by 2028 is a prediction about profiles, not a measured rate of deepfake interviews.
Updated 28 October 2026; added the 13 October 2026 DOJ action.
Key takeaways
- The FBI warned of deepfakes and stolen identities in remote-job applications on 28 Jun 2022.
- DOJ's 13 Oct 2026 announcement describes coordinated actions against North Korean remote IT worker schemes affecting 70+ US companies: 14 individuals charged, 2 arrests, nearly 190 devices seized. The defendants are charged, not convicted, and the release does not mention deepfakes or interviews.
- Gartner's 2Q25 survey (n=3,000) found 6% of candidates admit interview fraud. Its prediction that 1 in 4 candidate profiles will be fake by 2028 covers profiles generally, not video deepfakes.
- The employer-side figures collected on this page come from vendor-run surveys and vendor-reported data. Pindrop, for example, reported that 1 in 6 applicants showed signs of fraud for its own openings. The numbers disagree because they measure different things.
- On the sources collected here, most documented fraud is ordinary proxy and identity fraud. Documented deepfake use is rarer and concentrated in security research and DPRK-linked cases.
Three kinds of source get cited for deepfake candidates: government advisories, a Gartner prediction, and vendor or survey figures. They answer different questions, and a number quoted for one purpose does not transfer to another. This page separates them, with the date and primary source for each.
For a compact timeline, see the hub's documented interview-fraud events. This page carries the extended record and the comparison of methods.
The deepfake candidate record, 2022–2026
Every row links to its primary source where one was available. Rows are typed: (a) government finding, (b) court or DOJ allegation, (c) security-vendor research, (d) survey or prediction. The last column says what the row does not establish.
| Date | Event | Source | Type | What this does NOT show |
|---|---|---|---|---|
| 28 Jun 2022 | FBI IC3 PSA reports an increase in complaints about deepfakes and stolen personal information used to apply for remote jobs, and lists lip-sync and audio-visual mismatches as tells. | FBI IC3 | (a) | A rise in complaints is not a prevalence rate. |
| 21 Apr 2025 | Unit 42 (Palo Alto Networks): one researcher with no image-manipulation experience and a five-year-old computer builds a synthetic interview identity in 70 minutes. | Unit 42 | (c) | How often this happens in real hiring, or how often it passes a live interview. |
| 18 Jun 2025 | Pindrop reports that across its own remote openings, 1 in 6 applicants showed signs of fraud, 1 in 343 was linked to DPRK activity, and 1 in 4 of those used a deepfake in a live interview. | Pindrop | (c) | One employer's pool. Sample size: 541 applicants for one role, per the article. Pindrop sells tools that detect deepfakes. |
| 30 Jun 2025 | DOJ announces coordinated nationwide actions against North Korean remote IT worker schemes, including searches of 29 known or suspected laptop farms across 16 states, seizure of 29 financial accounts and 21 web domains, and more than 100 US companies affected. | DOJ | (b) | Allegations and enforcement actions, not convictions. The release does not establish that deepfakes were used. |
| 23 Jul 2025 | FBI IC3 PSA I-072325-4-PSA: US-based facilitators provide "attendance at virtual interviews and meetings on behalf of North Korean IT workers." | FBI IC3 | (a) | A facilitator attending is proxy attendance. The advisory does not count how many interviews were affected. |
| 31 Jul 2025 | Gartner press release: 6% of 3,000 candidates (2Q25) admit interview fraud; Gartner predicts 1 in 4 candidate profiles worldwide will be fake by 2028. | Gartner | (d) | The 1 in 4 is about profiles, not deepfakes. The 6% is self-reported. |
| 16 Sep 2025 | Checkr survey of 3,000 US hiring managers: 31% interviewed someone later found to be using a false identity. | Checkr | (d) | What it is NOT: verified incidents. It is manager recall in a vendor survey whose methodology is not detailed. |
| 13 Oct 2026 | DOJ: one indictment (N.D. Ohio) charges 14 individuals; 2 arrests; nearly 190 devices seized; 70+ US companies affected. | DOJ | (b) | Convictions. The defendants are charged, and the release does not mention deepfakes or interviews. |
Three rows concern deepfakes directly: the 2022 FBI warning, the Unit 42 build, and Pindrop's report from one employer's pool. None measures how often deepfakes occur across hiring. The DOJ and 2025 FBI rows describe people, laptops and facilitators. The survey rows measure self-admission and recall, not confirmed incidents.
What Gartner's "1 in 4" is and is not
Gartner's 31 Jul 2025 press release carries three numbers that tend to get blended:
- A survey of 3,000 candidates (2Q25): 6% admitted interview fraud, meaning posing as someone else or having someone else pose as them.
- A survey of 3,290 candidates (4Q24): 39% used AI in the application process.
- A prediction: by 2028, 1 in 4 candidate profiles worldwide will be fake. The release's wording is "Gartner predicts that by 2028, one in four candidate profiles worldwide will be fake" (Gartner, 31 July 2025).
The prediction is a forecast, not a measurement. "Fake profiles" also covers more than video: the figure says nothing about how many interviews involve synthetic media. Of the three numbers, only the 6% measures fraud, and it counts what candidates were willing to admit.
What DOJ and FBI have said
13 October 2026. The Justice Department announced coordinated actions against North Korean remote IT worker schemes affecting more than 70 US companies. One indictment in the Northern District of Ohio charges 14 individuals: two US nationals were arrested, ten China-based defendants are named, and two are sealed. The indictment alleges a scheme running from 2019 to July 2025 that generated more than $3 million. Separately, the FBI searched 40+ residences in Texas from 21 to 25 September ("Operation Lonestar Roundup") and seized about 187 laptops and KVM devices. CrowdStrike identified the group as "Famous Chollima." The alleged victims span health care, communications, cyber support, retail and manufacturing. Every one of these points is an allegation; the defendants are charged, not convicted.
The FBI Cyber Division described the outcome as "securing charges against 14 individuals, conducting two arrests, seizing nearly 190 devices." The Justice Department's statement adds that North Korean IT workers are increasingly recruiting American facilitators to counter improved private-sector detection, and in some cases subcontracting entire jobs to third-party foreign nationals. The release text does not mention deepfakes or interviews. The action cannot be cited as evidence of deepfake use. According to DOJ, the FBI issued a new PSA the same day.
According to the FBI's 2025 advisory, vendor-outsourced IT work can sit "removed from the direct hiring process." That is why outsourced IT work is a separate problem from interview deepfakes. See proxy fraud in the subcontracting chain. The IT services side is covered in proxy fraud in IT services hiring.
30 June 2025. An earlier DOJ action targeted laptop farms: searches of 21 premises across 14 states, with 29 financial accounts and 21 web domains seized and more than 100 US companies affected.
23 July 2025. FBI IC3 PSA I-072325-4-PSA says US-based facilitators provide "attendance at virtual interviews and meetings on behalf of North Korean IT workers." Its tips for interviewers include asking the person to wave a hand in front of the face, since this "may prompt a malfunction in AI generated video," and capturing images for later comparison.
28 June 2022. The FBI's public warning reported an increase in complaints about deepfakes and stolen personal information used to apply for remote jobs, and named lip-sync and audio-visual mismatches as tells.
How easy is a synthetic interview identity?
In April 2025, Evan Gordenker of Palo Alto Networks' Unit 42 reported that one researcher with no image-manipulation experience, limited deepfake knowledge and a five-year-old computer built a synthetic interview identity in 70 minutes. Unit 42 describes the analysis as consistent with the tactics, techniques and procedures of DPRK IT workers. It also tested interviewer probes, and found that asking the person to put a hand over their face was the most effective.
Treat the result as security-vendor research. The report does not say how often such an identity passes a live interview. This page does not reproduce the build method.
What do employer surveys show, and why don't they agree?
The figures differ because each source counts a different event. A candidate who admits fraud, a manager who remembers an incident, and an employer whose vendor flagged an applicant are three different things.
| Source | Date and sample | Who answered | Headline figure | Unit measured | What this does NOT show |
|---|---|---|---|---|---|
| Gartner | 31 Jul 2025; n=3,000 (2Q25) | Candidates | 6% admit interview fraud | Self-admission | Anyone who did not admit it; deepfakes specifically. |
| Checkr | 16 Sep 2025; n=3,000 | US hiring managers | 31% interviewed someone later found to be using a false identity; 19% extremely confident of catching fraud | Manager recall of past incidents | What it is NOT: verified incidents. Vendor survey; methodology not detailed. |
| Pindrop | 18 Jun 2025; 541 applicants for one role | Its own remote-role applicant pool | 1 in 6 applicants showed signs of fraud; 1 in 343 linked to DPRK; 1 in 4 of those used a deepfake in a live interview | One employer's pool, as classified by a vendor | Other employers' rates. Pindrop sells tools that detect deepfakes. |
| Greenhouse | 19 Nov 2025; U.S. n=665 recruiters and hiring managers | U.S. hiring managers | 18% caught applicants showing up as deepfakes | Deepfake encounters reported by managers | Verified counts. Vendor survey. |
Read across the rows, no source counts confirmed deepfake interviews. Gartner counts admissions, Checkr counts recollections, Pindrop counts one pool, and Greenhouse counts what US hiring managers say they caught. When two outlets cite different percentages, the first question is which of these units each one is using.
Keeping deepfakes in proportion
Most documented fraud on the sources collected here is ordinary proxy and identity fraud. DOJ's October 2026 text centres on stolen or fabricated documents, VPN and proxy infrastructure, facilitators and laptop farms. Documented deepfake use is rarer and sits mostly in security research and DPRK-linked cases. Putting deepfakes at the centre of the story overstates the record.
The target of every source above is a fraud operation. Ordinary candidates are not the story, and nothing in the record suggests they should be treated as suspects.
What hiring teams can do with this
Start by verifying identity at the first touch and comparing it at later stages; capturing images for later comparison is the FBI's own tip. A red-flag checklist mapped to verification steps turns that into a routine. Second, treat any flag as a trigger for a verification step, never as a verdict. Third, vary questions across rounds, so a prepared script or a synthetic persona has less to rehearse. The third point is our practice guidance, not a finding from the sources above.
Keep monitoring proportionate. EU AI Act Article 5(1)(f) bars AI inference of emotions in the workplace from 2 February 2025, which limits what an interview-monitoring tool may claim to read from a face or voice.
Real-time AI help during an interview is a separate problem from synthetic identity, and the sources behind it differ; see tools that coach candidates in real time. For buyers, the pillar question is the same across both: ask what a platform records and what it claims, not only what it markets. That is the core of choosing AI interview software.
FAQ
How many job candidates are fake?
There is no single measured rate. Gartner's 6% is self-reported (n=3,000, 2Q25), and its 1 in 4 is a prediction about candidate profiles by 2028. Surveys differ because they count different things.
Are deepfake job candidates real?
Yes, in the sense that they are documented in FBI warnings from 2022 and in security research such as Unit 42's 70-minute build. DOJ's October 2026 release does not mention deepfakes. The record establishes that deepfakes are possible and have been reported; it does not establish how common they are.
How do North Korean IT workers get hired?
According to DOJ's allegations, through fabricated documents, VPN and proxy infrastructure, American facilitators and laptop farms. These are allegations: the defendants are charged, not convicted.
What did the DOJ announce about North Korean remote IT workers?
On 13 October 2026, DOJ announced coordinated actions affecting more than 70 US companies. One indictment charges 14 individuals, two people were arrested, and nearly 190 devices were seized. The release does not mention deepfakes.
Can a deepfake pass a video interview?
Unit 42 showed that a synthetic interview identity can be built in 70 minutes. There is no pass-rate data in the record. Interviewer probes such as a hand over the face are recommended practice, not validated detection.
What does the FBI say about deepfake applicants?
In June 2022, the FBI reported more complaints about deepfakes and stolen personal information used in remote-job applications. In July 2025, it said facilitators attend virtual interviews on behalf of North Korean IT workers, and advised interviewers to capture images and try a hand wave in front of the face.
Why do the survey numbers disagree?
They measure different things: candidate self-admission, manager recall, and one employer's applicant pool. None is a count of confirmed deepfake interviews.
Is a deepfake the same as a proxy interview?
No. In a proxy interview, someone else answers; a deepfake is synthetic media. Most documented cases are proxy or identity fraud.
Next step
See what a ZeaHire evidence record contains: transcript, per-question scores and proctoring counters, for a human to review.
Comments
No comments yet — start the conversation.