How to Detect a Deepfake or Proxy Candidate
Remote hiring opened the door to a new kind of fraud: deepfaked faces, voice-swapped proxies, and applicants who are not who they claim to be. Here is how to catch it before an offer goes out.
The new shape of interview fraud
Interview fraud used to mean a padded resume. Now it can mean a real-time deepfake overlay, a hidden proxy answering questions off-screen, or a candidate feeding an AI assistant the questions and reading its answers. These schemes leave signals - in video liveness, audio timing, lip-sync, and answer latency - but only if you are looking for them and capturing the session.
- Deepfake video: subtle lip-sync drift, unnatural blink rate, edge artifacts
- Proxy interviews: eye movement off-screen, answer timing that lags the question
- AI-assisted answers: unnaturally polished latency and phrasing on hard prompts
- Identity mismatch: face or voice that shifts between sessions
What to check and when
| Stage | Signal to check | Response |
|---|---|---|
| Screening | Resume claims vs interview answers | Probe specifics live |
| Live interview | Video liveness and lip-sync | Automated deepfake check |
| Live interview | Answer latency on hard prompts | Ask an unscripted follow-up |
| Offer stage | Identity consistency across sessions | Verify identity documents |
A worked example
- 1Capture the interview session so signals can be analyzed, not just remembered.
- 2Run automated liveness and deepfake checks on the video and audio.
- 3Flag latency and phrasing anomalies on unscripted follow-ups.
- 4Escalate flagged sessions to a human recruiter for review.
- 5Verify identity documents before extending an offer to a flagged candidate.
How Fintra detects candidate fraud
Fintra runs candidate integrity and deepfake detection inside the same AI interview session that scores the candidate. It checks video liveness, flags audio and lip-sync anomalies, and surfaces latency patterns that suggest a proxy or AI assistant - then routes flagged sessions to a recruiter for a human decision. No separate tool, and every candidate is checked consistently.
- Deepfake and liveness detection on the live interview video
- Latency and phrasing anomaly flags for proxy or AI-assisted answers
- Flagged sessions routed to a human recruiter, never auto-rejected
- Consistent checks on every candidate, not just the suspicious ones
Frequently asked questions
How common are deepfake candidates?
Remote hiring fraud, including proxy interviews and AI-assisted answers, has risen sharply as the tools have become cheap and convincing. Any team hiring remotely at volume should assume it will encounter attempts and screen for them.
Can a human interviewer spot a deepfake?
Sometimes, but modern deepfakes are good enough to fool the eye, and humans only scrutinize candidates who seem off. Automated liveness and latency checks catch signals consistently across every candidate.
Should a flagged candidate be automatically rejected?
No. A flag is a signal, not a verdict. Route flagged sessions to a human recruiter who can run an unscripted follow-up and verify identity before making any decision.
What signals indicate a proxy interview?
Consistent answer latency before hard questions, eyes tracking off-screen, phrasing that reads as recited, and identity that shifts between sessions. Capturing the session lets you analyze these rather than rely on memory.
Stay in the loop
One practical finance briefing a week - new guides, checklists, and benchmarks.
Catch interview fraud before the offer
Deepfake, liveness, and proxy checks on every session.
Talk to us