Gartner projects that by 2028, one in four job candidates will use AI to misrepresent themselves during the hiring process. The projection, first released in late 2024 and updated in a January 2026 briefing, is not a fringe scenario. What breaks first isn't the technology — it's the screening infrastructure employers still rely on.
This guide is published by Mokka, an AI candidate screening platform. We include ourselves alongside competitors and aim to be accurate about both our strengths and limitations.
The baseline is already here. 54% of job seekers admitted to using AI tools on applications in a 2025 ResumeBuilder survey. The marginal cost of producing a flawless, customized resume has collapsed, and adoption has followed.
Free generation, evaporating stigma
A decade ago, tailoring an application to a specific job description took hours of human work. Today, it takes seconds of compute. The ResumeBuilder survey found that AI usage on applications jumped from roughly 30% in early 2024 to 54% by 2025 — a historical trajectory that shows the slope of adoption. Adoption curves don't climb that fast when a tool requires capital or specialized knowledge.
"We're past the tipping point. Candidates don't even view AI assistance as cheating anymore — it's seen as leveling the playing field."
Stacie Haller, Chief Career Advisor at ResumeBuilder
When candidates perceive AI assistance as a legitimate way to compete against employer-side automated sorting, social stigma evaporates. The candidate isn't sneaking around; they're adapting to an environment where their application is already being evaluated by an algorithm. This cultural normalization makes the problem far harder to address through policy or threats.
What 'fake' actually means
The word "fake" is doing too much work in most discussions of this trend. Gartner's projection covers a spectrum of AI-assisted misrepresentation, and conflating these distinct behaviors leads to bad screening design.
The spectrum runs from cosmetic enhancement to identity fraud. Cosmetic enhancement — AI-polished language that makes a mediocre candidate sound exceptional on paper — inflates false positives in resume screening but doesn't necessarily involve fabricated credentials. Capability inflation sits in the middle: claiming proficiency in skills the candidate doesn't have, with AI generating plausible technical narratives to back it up. Identity fraud is the extreme — a different person entirely taking the interview, sometimes using deepfake technology or real-time AI prompting to simulate expertise they don't possess.
The HireRight 2025 Employment Screening Benchmarking Report found that candidate fraud incidents reported by employers increased 92% from a 2023 baseline to 2025. But that headline number masks the breakdown. Identity-level fraud is far more concentrated in remote-first roles, which see 2.3x more instances of identity misrepresentation compared to on-site positions, according to Sterling's 2025 screening report. Cosmetic enhancement is everywhere; identity fraud concentrates where verification is weakest.
Built for paper, stressed by code
The screening stack most companies use was built for assumptions that no longer hold. Josh Bersin called it bluntly in 2025: the infrastructure was built for the paper resume era, and it was already creaking before ChatGPT arrived.
The standard funnel — resume screening, phone screen, video interview, background check — was designed around the assumption that the person applying is the person who will show up and do the work. Dr. John Zoldak, CEO of Veriff, put it precisely in Q1 2026: "Traditional background checks were designed for a world where the person applying was the person showing up. That assumption is now broken at scale."
The breakage happens at specific stress points. Resume screening floods with false positives when every applicant can generate a perfectly optimized document. Take-home assessments lose validity when a candidate can paste the prompt into an AI and receive a polished deliverable in minutes. Asynchronous video interviews — recorded responses to standard questions — can be scripted and telepromptered. Even live video interviews are compromised: 38% of employers reported catching candidates using deepfake technology or AI voice tools during video interviews in 2025-2026, per iProov's 2025 identity fraud study. The February 2026 case of a candidate caught using real-time AI prompting during a live technical interview at a FAANG company was notable only because the detection succeeded.
Meanwhile, the cost of failure compounds. SHRM's 2025 data puts average cost per hire at $5,400, and re-hires due to fraudulent candidates cost three times the original hire expense. If your screening process lets through even a small percentage of capability-inflated candidates, the downstream cost dwarfs the investment required to fix the funnel.
The no-win recruiter dynamic
Recruiters are caught in a structural trap. Tighten screening, and you extend time-to-hire, add friction to the candidate experience, and risk losing strong candidates who won't tolerate a gauntlet of verification steps. Keep current speed, and you absorb higher bad-hire rates, which are invisible in real-time but devastating on a six-month look back.
72% of TA leaders say AI-generated resumes and cover letters make it significantly harder to assess authentic candidate capability, according to LinkedIn Talent Solutions data from Q1 2026. The problem isn't effort or skill — it's tooling. You cannot human-speed your way out of a machine-speed problem.
The legal layer makes it worse. EEOC issued informal guidance in November 2025 warning that AI-based screening tools must not create disparate impact when flagging "suspicious" applications. The compliance concern is legitimate, aggressive fraud detection can disproportionately flag non-native English speakers, neurodivergent candidates, or other protected groups whose communication patterns deviate from a narrow norm. But the practical effect is that legal teams block aggressive detection, leaving TA teams without workable tools.
The detection problem is an arms race
LinkedIn's March 2026 launch of real-time AI-detection features for job applications was the first major platform-level response to candidate fraud at scale. The feature flags AI-generated content as it's submitted. It's a meaningful step, but it's also an escalation in an arms race that employers can't win on the detection side alone.
Here's why: AI detectors suffer from two compounding problems. First, false positives destroy candidate experience and create legal exposure. Second, the generation technology improves faster than detection technology. A detector trained on GPT-3.5 output is already obsolete; models generate text that's statistically indistinguishable from human writing within months.
HireVue's October 2025 study quantified the gap: 63% of candidates who use AI on applications still pass initial screening. Current filters aren't catching sophisticated use. And as Madeline Laurano of Aptitude Research noted in Q4 2025, the real fraud rate may already exceed 1 in 4 for certain segments, tech roles, remote-first positions, and high-volume hiring see disproportionate rates.
What actually works: assessment-first hiring
The employers making real progress on this problem aren't building better detectors. They're redesigning the hiring process to make detection less necessary.
The shift is from credential signaling to demonstrated capability. Instead of screening on resumes and then verifying at the end, you assess capability early and use credentials as a secondary signal. This inverts the funnel.
A framework for thinking about this: verification should move to the point of action. Don't verify identity at onboarding, verify it at the interview. Don't assess coding skill via take-home, assess it live, in a collaborative environment where real-time AI use is either impossible or irrelevant because the assessment is designed around interaction, not output. Several enterprises began piloting in-person verification checkpoints for remote hires in April 2026, per Wall Street Journal reporting, reversing years of fully virtual onboarding. The pendulum is swinging back toward physical presence as a verification mechanism.
The deeper structural answer is that screening needs to shift from what you've done (which can be fabricated) to what you can do right now, in this room, with these constraints. That's a fundamentally different process design, and it requires different tooling.
The 78% problem
Only 22% of organizations have updated their screening processes to account for AI-assisted fraud as of mid-2026, according to Gartner's Q1 2026 HR survey. That means 78% of employers are running 2026 volume through a screening stack designed for 2019 assumptions.
Gartner's Emily Rose McLaughlin argued that the "1 in 4" figure isn't hyperbole, it's a structural consequence of free or near-free generative AI colliding with remote hiring at scale. The projection may be conservative for the segments where fraud concentrates.
The mental model: shift-left verification
The framework I use to think about this borrows from software engineering. In security, there's a principle called "shift-left testing", move quality checks as early as possible in the development cycle, because bugs found late cost orders of magnitude more to fix.
The same principle applies to hiring fraud. Shift-left verification means moving identity and capability checks to the earliest possible stage of the funnel, before you've invested recruiter time, before you've extended an offer, before the bad hire is sitting in a seat costing you three times their salary to replace.
This is where the architecture of modern hiring platforms matters. At Mokka, the approach is to build verification into the sourcing and screening process itself rather than bolting it on as a final checkpoint. When a sourcing agent identifies a candidate, the AI Evaluation Agent, which screens resumes and conducts AI pre-interviews, can immediately pair that with a live capability assessment and identity verification, before a human recruiter ever spends time on the profile. The cost structure flips: instead of spending recruiter hours on candidates who may not be who they claim to be, you spend compute on verification and reserve human attention for verified, assessed candidates.
Mokka has real limitations. We're an early-stage company, our ATS integrations are still limited during pilot deployments, our seat-based pricing gets expensive for large enterprise teams, and we're not built for executive search. What we are built for is high-volume hiring where verification-at-the-point-of-assessment matters most.