Most writing on this subject is either alarmism or denial. This page is the evidence base instead: what has been measured, by whom, when, and what it does not yet tell us. Every figure below is attributed and linked, including the ones that weaken the case.
Two years ago, an employer worried about interview cheating was worried about a browser tab. The concern now is categorically different, and it moved faster than most hiring processes did. These four figures, from four independent sources, are the ones worth knowing.
Read the second and fourth together, because that is where the story actually is. Six percent of candidates admit to impersonation in an anonymous survey, which is already a remarkable number for a self-reported crime. Forty-one percent of large employers say a fraudulent candidate got all the way through to onboarding. The detection gap between those two figures is the entire problem.
The most useful survey on this subject did not ask recruiters. It asked 668 IT, cybersecurity, fraud and risk leaders at organisations with more than 1,000 employees, across 15 industries, in September 2025. These are the people who find out what actually happened after a hire goes wrong.
Almost every confused argument about interview cheating comes from treating it as a single behaviour. It is two, and they need completely different responses.
A real person, applying for a real job they may even be qualified for, who uses an assistant during the interview because the tool exists, it is cheap, and they believe everyone else is using one. This is the overwhelming majority by volume. Their behaviour is elastic: it responds to whether they think they will be caught, which is why disclosure and deterrence work on this group and why heavy-handed surveillance is unnecessary for it.
Someone who is not the person on the resume at all. A proxy sitting the interview for a paying client, a candidate whose identity is synthetic, or an organised operation placing workers into salaried roles. This group is far smaller and far more expensive to get wrong, because the consequence is not a bad hire. It is an unknown person with your credentials.
"Candidate fraud creates cybersecurity risks that can be far more serious than making a bad hire."
JAMIE KOHN, SENIOR RESEARCH DIRECTOR, GARTNERConfusing the two leads to both classic failures. Treat everyone as an impostor and you build an interrogation that drives away good candidates. Treat everyone as an opportunist and you never catch the case that actually costs you something.
The clearest documented example is not a survey. It is a United States Department of Justice prosecution, which means the facts were established in court rather than self-reported.
From roughly 2021 until October 2024, a scheme placed remote IT workers into jobs at more than 100 US companies, many of them Fortune 500. The workers passed the interviews. The identities were real people's, stolen. Company laptops were shipped to residences inside the US and wired to remote-access switches, so the work appeared to be happening from an American home office while the person doing it was somewhere else entirely.
Two details in that case matter more than the headline numbers. The first is that these hires passed normal interviews at sophisticated companies, repeatedly, for three years. The second is the remote-access switch. Every one of those workers was operating a machine they were not sitting in front of, which is a signal that exists at the operating-system level and nowhere else. It is invisible on a video call, invisible on a screen share, and invisible to a background check, because the background check was run on a real person whose identity had been stolen.
The widely cited global figure is that a bad hire costs around 30% of that employee's first-year salary. That number is old, conservative, and drawn from a labour-market context that is not India's. Indian analyses put the real figure considerably higher, for three structural reasons: notice periods of 60 to 90 days, a full re-recruitment cycle rather than a quick backfill, and the disruption cost at senior levels.
| Component | Typical basis | Mid-level hire at ₹20 LPA |
|---|---|---|
| Salary paid during poor performance | around 5 months | ₹8,30,000 |
| Recruitment agency fee | 10 to 15% of annual CTC | ₹2,00,000 |
| Lost productivity | roughly 30% of expected output | ₹2,50,000 |
| Re-recruitment, full cycle again | repeat of the original spend | ₹2,00,000 |
| Manager time and team disruption | hardest to measure, rarely zero | ₹2,00,000 |
| Indicative total | roughly 90 to 100% of first-year CTC | ₹16,80,000 |
The India-specific percentages here come from industry analysis rather than peer-reviewed research, and the components are estimates rather than audited costs. Treat the structure as sound and the precise total as indicative. The reason it is worth showing at all is the order of magnitude: one bad hire at a mid-level salary costs more than most companies spend on hiring tools in a year.
And that is the cost of an ordinary bad hire, where the person simply could not do the job. It does not include the cases that appear in the section above, where the cost is a credential handed to someone whose identity you never actually established.
Point predictions about 2028 are guesses. The trend underneath them is not, and it has been remarkably consistent: each generation of interview cheating moves further away from the browser and deeper into the machine.
The next step is visible from here. Assistants that listen to the question and produce the answer without the candidate acting as an intermediary at all, leaving a human present as a face and a voice while something else does the thinking.
What that trend implies for buyers is narrower than it first appears. Any defence that operates inside the browser, or inside the video call, is watching a layer the problem left several years ago. That is not a criticism of any particular product. It is a structural limit on where a browser extension or a video-analysis model can see.
All figures were checked against their primary source on 22 August 2026. Where a figure comes from vendor research rather than independent study, it is labelled as such.
WatchHire watches for the signals described on this page, at the layer they occur on, and writes your team a plain-English report the moment the call ends.