Deepfake candidates and interview fraud

By SkavanderHR · 13 September 2026

Interview fraud stopped being hypothetical. Gartner found 6% of about 3,000 surveyed candidates admitting to it, and forecasts that one in four candidates worldwide will be fake by 2028. Where it thrives is a large pool of strangers who arrived anonymously for a remote role. That is a description of a job posting, and it is worth knowing what changes when candidates do not arrive that way.

Key takeaways

• Measured, not predicted: 6% of about 3,000 candidates surveyed by Gartner admitted to interview fraud, such as having someone else pose as them.
• Predicted, and labelled as such: Gartner forecasts one in four candidates worldwide will be fake by 2028.
• 41% of large organisations report having already onboarded an impersonator, and setting up a convincing deepfake takes about 70 minutes.
• Once detection tools are in use, 25% to 30% of flagged sessions turn out to be fraudulent.
• Approaching people yourself does not detect fraud. It removes the conditions fraud needs, which is a different and more durable kind of protection.

How common is interview fraud, really?

Reported application and interview fraud figures

Figure	Measured or predicted	Source
6% of about 3,000 surveyed candidates admitted interview fraud	Measured, self-reported survey	Gartner, July 2025
One in four candidates worldwide fake by 2028	Predicted, explicitly a forecast	Gartner, via HR Dive
41% of large organisations have onboarded an impersonator	Measured, self-reported, US-centric	Industry coverage, 2026
About 70 minutes to set up a convincing deepfake	Measured, practitioner estimate	Industry coverage, 2026
25% to 30% of flagged sessions prove fraudulent	Measured, only where detection is used	Industry coverage, 2026
28% of job seekers admit generating fake work samples	Measured, self-reported	Greenhouse, 2025
32% claimed skills in AI they do not have	Measured, self-reported	Greenhouse, 2025

Sources: Gartner’s July 2025 candidate survey and its own forecast as reported by HR Dive; Greenhouse’s job seeker survey for the self-reported figures; the impersonator, setup time and flagged session figures come from the 2026 trade coverage of interview fraud recorded in our topic research, and that market data is largely United States based. Every figure here except the Gartner forecast is self-reported by candidates or employers, which typically understates rather than overstates the behaviour, and the flagged session rate only exists in organisations that already screen for it.

Two of these need reading carefully. The one in four figure is a forecast by analysts, not a count of anything, and it should never be quoted as a current measurement. And the flagged session rate is conditional: it says what detection finds among sessions it suspected, not what share of all candidates is fraudulent.

Why do anonymous applicant pools make this easy?

Because the whole arrangement is built on documents the candidate produced and a video call with a stranger. At an average of 254 applicants per posting, nobody is verifying identity on the way in, and there is no prior relationship against which anything is checked. The candidate chose which employer to approach, which story to tell, and which evidence to submit. Fraud only has to survive a process designed to filter for skill, not to authenticate a person.

Remote work widened this without anyone deciding to. When the whole process from application to onboarding happens without a room, the moment where an inconsistency would usually surface never arrives. And the trade press is not treating this as a fringe risk: fraudulent and AI-assisted candidates were ranked the most anticipated hiring challenge of the year, ahead of talent shortage.

Does an approached candidate fix it?

No, and it matters to say so plainly. There is no sourcing method that detects a deepfake, and anyone claiming immunity is selling you something. Verification is its own discipline: identity checks, a live technical conversation with follow-up questions that cannot be scripted, references you sought rather than ones you were given.

What changes with an approached candidate is the exposure, and it changes in three specific ways.

The direction of contact reverses. You found the person because of work that already existed in public under their name, with dates, collaborators and a history that predates any interest in your role. Nothing about that record was assembled for your process.

The population shrinks to people who exist independently of your posting. A fraudulent candidate needs to be discoverable in order to be approached at all, which is a much harder bar than being able to submit a form.

And the numbers fall to a scale a human can actually check. Verifying a handful of people properly is achievable. Verifying hundreds is not, which is why the volume channel is where this lands hardest.

What this means for a company of twenty to five hundred people

Dedicated fraud detection is an enterprise purchase with an enterprise team behind it, and at this size neither is usually on the table. Three things are free.

Ask about work you found yourself rather than work that was submitted to you, and go a layer deeper than the summary: a decision inside a project, why one approach was dropped, what broke. A prepared artefact rarely survives a specific follow-up question.

Keep one video conversation with a real back and forth in it, rather than a set of recorded answers. Recorded, one-way stages are the easiest part of the process to fake and the hardest to interrogate.

And treat the first point of contact as part of the control. A candidate you approached, whose public record you read before they knew the role existed, is a different starting position from an unknown submission, and it costs nothing to prefer it. When you are weighing an agency, an embedded recruiter and a standing monthly search, that difference belongs in the comparison.

Where our part sits

We find and engage people, and we do not screen or authenticate anyone. For one specific role, the system looks through professional networks and company websites for the people in your market whose actual work matches the criteria, and each of them is told about the role about that role. Your own team takes every interview and makes every judgement, including the ones about identity, on a short list of people with a public work history you can read for yourself. Everyone relevant is usually reached within two weeks, and it costs about a tenth of a recruiting agency. For a family like security engineering, where candidates are assessed by people who know exactly how this is done, the sourced route is the one your own engineers will trust.

Fraud follows volume and anonymity. Remove both and you have removed most of the opportunity, without pretending you removed the risk.

Numbers from our own searches. Written and checked by SkavanderHR.