Overview
- Generative AI lets anyone produce a polished CV, a fake work history or a deepfake interview in minutes
- Gartner predicts that one in four candidate profiles worldwide will be fake by 2028
- In South Africa, 8.5 million unemployed people and a shift to remote hiring make the volume problem worse
- Verification now belongs at the top of the hiring funnel, not at offer stage
What does candidate fraud look like in 2026?
Candidate fraud is any attempt to misrepresent who a candidate is, or what they can do, in order to get hired. In 2026 it goes well beyond a stretched job title.
- AI-written CVs tuned to your job spec, with invented employers and fabricated results
- Proxy interviews, where someone else answers on the candidate's behalf
- Deepfake video, where AI overlays a synthetic face or voice onto a live call
- Prompt injection, where hidden text in a CV instructs an AI screener to rank the applicant highly
The last one is new, and it targets your software rather than your recruiters. Any team that uses AI to screen CVs needs a way to catch it.
Why has hiring shifted from scarcity to verification?
The bottleneck in recruitment is no longer finding candidates. It is trusting the ones already in your talent pipeline.
Generative AI has cut the cost of applying to almost nothing. Application volumes have climbed while the share that survive scrutiny has fallen. Research from both sides of the market shows the same pattern:
- Gartner found that 6% of candidates admit to interview fraud, and predicts one in four candidate profiles worldwide will be fake by 2028
- Greenhouse's 2025 AI in Hiring Report found that 91% of US recruiters have spotted candidate deception, and 34% spend up to half their week filtering junk applications
- In the same study, 41% of US job seekers admitted to using prompt injection, and 18% of hiring managers had caught a deepfake
When a third of a recruiter's week goes on filtering fakes, the constraint is trust, not supply.
Why is South Africa more exposed to candidate fraud?
South Africa combines extreme application volumes with growing remote hiring, which gives fraud more cover than in most markets.
Volume. Stats SA's Quarterly Labour Force Survey for Q2 2026 put official unemployment at 33.6%, with 8.5 million people out of work and youth unemployment at 47.4%. A single entry-level or learnership advert can attract thousands of applications. AI-generated CVs push that number higher and make each one harder to tell apart.
Remote hiring. When a candidate can be sourced, screened, interviewed and onboarded without entering a building, a borrowed or invented identity is far harder to catch.
Compliance data. Employment Equity reporting and POPIA records assume candidates supplied real information. Fake profiles quietly distort the applicant data your EE plans and audits depend on.
Why don't background checks catch candidate fraud?
Background checks run too late and verify the wrong thing. They happen after the offer, by which time a fraudulent candidate has already used up screening and interview time.
They also check the details the candidate supplied. A stolen identity with a clean record passes, because the check confirms the record, not that the person on the call owns it.
Gartner's own recommendation is to embed identity verification and anomaly alerts inside recruiting systems, rather than relying on a final check. That means moving verification forward.
How do you move verification to the top of the funnel?
Run light checks on every applicant early, and save heavier checks for the shortlist. Each stage filters out a different kind of fraud.
|
Stage |
Check |
What it catches |
|---|---|---|
|
Application |
Duplicate and pattern checks: recycled CVs, identical phrasing, bursts of applications from one contact number |
Bots and CV farms |
|
Screening |
The same structured, role-specific questions for every applicant |
Invented experience that falls apart under specific questions |
|
First interview |
Camera on, photo ID matched to the face |
Proxy candidates and borrowed identities |
|
Shortlist |
Unscripted follow-ups, a short skills task, SAQA qualification checks |
Deepfakes, scripted AI answers, fake qualifications |
|
Offer and day one |
Confirm the person starting is the person who interviewed |
Bait-and-switch hires |
The first two stages cost almost nothing per applicant, which is why they can run at volume. Technology carries most of that load:
- Standardise screening. Chat-based screening through txthr asks every applicant the same questions on WhatsApp or SMS, so answers that don't fit stand out.
- Keep one auditable record. Neptune ATS holds every application, screening answer and decision in one place, which is what makes recycled CVs and odd application patterns visible.
- Test capability, not CV writing. Skills-first screening is much harder to fake than a keyword match.
- Automate continuity checks. A fraud detection agent can confirm identity across stages and flag scripted answers for review.
- Weight trusted sources. Employee referrals and internal talent pools carry far less fraud risk than cold applications.
How do early identity checks fit with POPIA?
Early identity checks are compatible with POPIA if they are proportionate, explained and consented to. The detail that matters is biometrics.
Face matching processes biometric data, which POPIA classes as special personal information. That calls for specific, informed consent and a clear purpose. A photo ID held up to the camera is a far lighter step than an automated face match.
The practical answer is tiered checks. Use light checks for everyone, and reserve biometric checks for the shortlist or for roles with access to systems, money or sensitive data. A recruitment compliance workflow should record what was checked, when, and on what basis.
Should recruiters trade speed for verification?
Yes, but only at the top of the funnel. Our view is that time-to-hire is the wrong headline metric when a quarter of profiles could be fake.
A fast process that hires a proxy candidate hands a stranger access to your systems, data and payroll. Gartner warns that the cybersecurity exposure can outweigh the cost of a bad hire. Talent teams should report a verified-candidate rate, the share of shortlisted candidates who pass identity and qualification checks, alongside time-to-hire.
The opposite mistake is overcorrecting. Mandatory in-person interviews for every role are not realistic in high-volume South African hiring. Layered, proportionate checks are, provided a human still makes the final call. An unscripted follow-up question remains one of the best fraud detectors available.
Frequently asked questions
How common is candidate fraud?
Gartner predicts that one in four candidate profiles worldwide will be fake by 2028, and 6% of candidates already admit to interview fraud. Greenhouse found that 91% of US recruiters have spotted candidate deception.
How can recruiters spot a proxy or deepfake interview?
Ask for the camera to stay on and match the face to photo ID at the start. Then ask unscripted follow-up questions. Lips out of sync with audio, a face that glitches when a hand passes in front of it, and answers that sound read aloud are common warning signs.
Are early identity checks allowed under POPIA?
Yes, provided they are proportionate, explained to candidates and have a lawful basis. Biometric checks such as face matching involve special personal information, so they need the candidate's specific, informed consent.