Hiring teams built AI screening tools to make candidate evaluation faster and more consistent. Instead, the same technology has handed candidates a faster, more consistent way to fake their way through the process, and new research suggests employers are losing the arms race.
The scale of the problem
A 2026 research report from AI-proctoring firm Aiseptor, which compiles fraud data across assessment platforms, hiring vendors, and academic literature, found that cheating attempts on technical hiring assessments have roughly doubled in a year, from 16% of candidates in 2024 to 35% in 2025. On unproctored take-home tests, the kind most remote-first employers rely on to keep hiring fast, Aiseptor’s analysis found scores run about four times higher than the same candidates produce under proctored conditions.
The detection gap is the more uncomfortable number. Aiseptor’s research puts the undetected rate at 61%: roughly six in ten flagged cheating attempts still clear the hiring bar because most employers have no behavioral-detection layer sitting behind their assessment tools. Fraud-related hiring losses rose at 60% of companies between 2024 and 2025, according to the same analysis.
Why candidates are doing it
The report’s most telling figure is not about capability, it is about intent. Only 14% of candidates openly admit to using AI to misrepresent their skills, but 83% say they would do it if they believed they would not get caught. Aiseptor’s researchers put the true usage rate closer to 35% to 42%, arguing that perceived risk, not ethics, is what is actually holding the line. Fifty-nine percent of hiring managers now say they suspect candidates of using AI during live assessments, which means the suspicion has already outpaced the tooling built to confirm it.
The economics make the behavior rational rather than reckless. AI cheating tools cost candidates $20 to $50 a month against starting salaries of $120,000 to $180,000 for the technical roles most targeted. Entry-level and campus-recruiting pipelines see the highest rates, Aiseptor found, because they combine the strongest financial incentive with the weakest verification.
Where the fraud concentrates
The rate is not evenly spread. Aiseptor’s analysis puts technical and engineering roles at a 48% cheating rate on assessments, more than double the 12% rate seen in sales and other non-technical roles, and finds junior candidates cheat at roughly twice the rate of senior ones. Geography matters too: assessment fraud attempts run around 48% in Asia-Pacific hiring pipelines against roughly 27% in North America, a gap the researchers tie to the same salary-differential dynamics that make cheating economically attractive for entry-level US roles. Detection tools carry their own bias problem on top of this: Aiseptor’s report cites a 61.2% false-positive rate from AI writing-detection tools when flagging non-native English speakers, compared with 5.1% for native speakers, meaning a blunt reliance on AI-detection software risks screening out legitimate international candidates at a far higher rate than it catches actual fraud.
Remote hiring made it worse
Remote-first recruiting removed the proctored room, and with it, most of the friction that used to make impersonation and real-time coaching difficult. Aiseptor’s report describes candidates receiving live AI assistance during video interviews and, in more organized cases, having a more qualified person sit in for the actual interview. Neither behavior leaves the kind of behavioral fingerprint that traditional plagiarism or proctoring software was built to catch, which is why detection has lagged so far behind adoption.
What it means for the HR leader
The cost of getting this wrong is not abstract. Aiseptor’s analysis puts the cost of a single bad technical hire at $42,000 to $125,000 for junior roles and $200,000 to $750,000 or more for senior and VP-level hires, once ramp time, team disruption, and re-hiring are factored in. That is the number that should be driving budget conversations, not the cost of a proctoring tool.
A few practical shifts follow from the data:
- Treat unproctored assessments as a screen, not a verdict. A four-times score inflation gap means an unproctored result should narrow a pool, not clear a candidate for an offer.
- Add a live, unscripted component before final offers for roles where a bad hire is expensive to unwind. Live interaction remains the hardest thing for AI-assisted cheating to fake convincingly in real time.
- Reset expectations with hiring managers. With 59% of managers already suspicious of candidates, the risk is that they start discounting every strong assessment score, including legitimate ones. A clear, communicated verification policy prevents that erosion of trust from spreading to honest applicants.
- Watch the vendor response, not just the candidate behavior. Aiseptor’s report notes that major employers are already moving toward in-person interviews and candidate attestations as countermeasures. HR technology buyers should expect their applicant tracking and assessment vendors to start shipping detection layers as a baseline feature rather than an add-on.
The bottom line
This is not a story about a handful of dishonest applicants. It is a story about a verification gap that widened faster than most hiring stacks were built to handle, on infrastructure companies like Workday and other identity-verification vendors are only now racing to close. Employers who treat AI-assisted cheating as a niche integrity issue rather than a structural hiring-process problem are the ones most likely to be explaining an expensive bad hire to their CFO next quarter. The legal exposure is rising in parallel: hiring tools themselves are increasingly becoming named defendants when their screening decisions go wrong, which raises the stakes for getting verification right on both sides of the process.
The near-term fix is not more AI, it is more friction at the moments that matter: live conversation, verified identity, and a hiring process that treats a clean assessment score as a starting point rather than a finish line.
Source: Aiseptor
