How to Stop Proxy & Deepfake Candidates in Interviews (2026)
Proxy candidates, AI-assisted answers and deepfakes are beaten the same way: continuous identity + liveness checks, a spoken interview that's hard to fake, and AI-assistance detection — not a one-time face scan at login.
Interview fraud isn't a fringe problem anymore — in remote and high-volume hiring it's routine. The good news: proxy candidates, mid-call ChatGPT use, and deepfakes all fail against the same defence. It's not a better one-time face scan; it's continuous verification plus an interview that's genuinely hard to fake.
The three ways candidates fake an interview
- Proxy / impersonation — a more skilled person takes the interview in the candidate's place.
- AI-assisted answers — the real candidate reads answers from ChatGPT or a co-pilot mid-interview.
- Deepfake / replay — a synthetic or pre-recorded video stands in for a live human.
Why one-time face verification fails
Most tools check identity once, at login — and then never again. That catches almost nothing, because the swap or the assistance happens after the check. In our own dataset of confirmed cheating cases, the large majority had passed initial face verification cleanly; continuous monitoring is what caught them. (More in our research on AI interview cheating in India.)
What actually stops it
Layer these — no single signal is enough on its own:
- Continuous face + gaze monitoring throughout the interview, not just at the start.
- Liveness challenges — random blink/head-turn prompts that defeat replays and most real-time deepfakes.
- A spoken, adaptive interview — unscripted follow-ups are hard to outsource or read from a screen.
- AI-assistance detection — timing, filler-word and vocabulary signals that flag mid-call chatbot use.
- Browser-side biometrics — process face data on the candidate's device so it's privacy-first.
Why voice interviews raise the bar
A live AI voice interview is a two-way conversation that adapts to each answer. You can't pre-record it, and reading from a chatbot introduces exactly the timing and phrasing anomalies that detection looks for. Paired with continuous proctoring, it closes most of the gaps that one-way video leaves open — see how this compares across tools in our guide to the best AI interview software.
A practical checklist
- Verify identity continuously, not once.
- Use liveness challenges at random intervals.
- Prefer a live, adaptive (ideally voice) interview over one-way recordings.
- Turn on AI-assistance detection and review flagged sessions.
- Keep biometric processing on-device for privacy and compliance.
Proof, not promises: the Interview Integrity Certificate
Detection only matters if you can show the result. Every interview on Interloop now ships with a verifiable integrity certificate: a 0–100 session-integrity score aggregated from continuous face match, liveness, environment monitoring (tab switches, paste events, fullscreen exits) and AI-assistance analysis of the transcript — published at a unique QR-verifiable link.
Share it with hiring managers, clients, or auditors: anyone can scan the QR and confirm, live against Interloop's records, that this person, not a proxy, completed this interview, without AI-written answers. If a forwarded report doesn't match the live certificate page, it's fake.
Frequently asked
What is a proxy interview?
A proxy interview is when someone other than the actual candidate takes the interview — a more skilled stand-in answering on their behalf. It's a major problem in remote and high-volume hiring, especially in IT and BPO.
Can AI detect ChatGPT use during an interview?
Yes — to a degree. Response-timing patterns, sudden vocabulary shifts, and unnatural pauses are detectable signals of mid-interview AI assistance. Combined with a live, adaptive conversation (where the AI asks unscripted follow-ups), it becomes much harder to lean on a chatbot undetected.
Do deepfakes work against a live interview?
Real-time deepfakes are getting better, but they struggle against continuous liveness challenges (random blink/head-turn prompts), gaze tracking, and a natural back-and-forth conversation. The defence is continuous verification, not a single check at the start.
Is continuous proctoring privacy-invasive?
It doesn't have to be. The strongest approach processes face data in the candidate's browser so biometrics never leave their device, while still flagging identity and liveness anomalies to the recruiter. That keeps proctoring effective and privacy-first.
Get the weekly hiring-intelligence brief
3-min read, every Tuesday. Real data, zero spam — unsubscribe with one click.
See an AI voice interview live.
Run a free voice interview in your candidate's language — and watch the proctoring and scoring work in real time.