After interviewing dozens of fraudulent candidates firsthand, it became clear that this is not one phenomenon. It’s a family of behaviors with different tools, motivations, and signals.
There are three dominant patterns.
1. Identity Fraud: The Ghost Applicant
This is the most traditional form of deception, and still shockingly effective.
The candidate on paper does not exist. The resume is fabricated. Credentials are fake or stolen. The interviewee may be a proxy, often operating from overseas, posing as the person on the resume.
Sometimes this is solo opportunism: get hired, collect paychecks until fired, steal equipment, move on.
Sometimes it’s organized, including state-sponsored labor laundering where U.S. companies unknowingly fund hostile regimes through fake hires. Federal investigations have confirmed cases where North Korean operatives took remote U.S. engineering jobs using stolen identities and worked through intermediary “laptop farms” hosted by American accomplices.
In interviews, identity fraud shows up through subtle breakdowns:
- Voices and accents that don’t match claimed educational histories
- Answers that sound processed rather than spontaneous
- Personal facts that collapse when drilled into (campus buildings that no longer exist, projects that were never real) Inconsistent digital trail across platforms
It isn’t that these candidates forget lies. It’s that their stories weren’t lived.
2. Real-Time AI Cheating: The Invisible Script
These candidates are real humans. But the intelligence driving the interview isn’t.
In roughly one-third of interviews, we now detect signs that candidates are using AI assistance in real time. Tools like ChatGPT, Claude, and specialized interview cheating software now
operate invisibly during video calls. They can generate adaptive responses, code snippets, and system design explanations fast enough that the candidate simply reads.
What makes this dangerous is not that candidates supplement their thinking. It’s that interviews no longer measure thinking at all.
These interviews sound impressive:
- Answers are unusually flawless
- Structure is polished
- Vocabulary is elevated
- Timing is unnaturally smooth
But real cognition has texture. Human thought includes uncertainty, hesitation, calibration, and self-correction. AI-assisted candidates often lack all of them. Their answers sound like writing, not speaking.
Interviewers mistake fluency for understanding. It’s not the candidate’s fault alone. The interview itself is no longer aligned to reality.
3. Synthetic Candidates: The Deepfake Interview
This one feels like science fiction until you watch it carefully.
The person on screen is not a real human. It is an AI-generated face, injected into the video feed using camera-spoofing software.
Most recruiters cannot reliably detect deepfakes. When we test hiring managers with real and synthetic candidates side-by-side, accuracy rarely exceeds 20%.
The tells are subtle:
- The edges of the face blur under motion
- The background behaves unnaturally
- Facial expressions lag speech
- Hand movements break depth perception
- Emotion appears scripted rather than responsive
How Organizations Are Preventing Interview Fraud in Remote Hiring
As interview fraud techniques continue to evolve, many organizations are introducing dedicated integrity layers within their hiring processes to address growing concerns around AI fraud and deepfake-based manipulation. Modern tools like Sherlock AI help recruiters maintain trust in remote interviews by detecting signals that may indicate external assistance or suspicious activity. Instead of relying solely on manual observation, these platforms analyze behavioral cues and interview patterns to highlight potential risks such as AI-generated responses, AI-assisted cheating, proxy candidates, or identity inconsistencies that may be linked to deepfake technology. This added visibility allows hiring teams to evaluate candidates with greater confidence, even in fully remote interview environments.
Platforms like Sherlock AI focus specifically on protecting the integrity of live interviews by identifying AI copilots, hidden real-time assistance tools, and unusual response patterns that may suggest AI-assisted cheating. The system can also flag potential impersonation attempts and deepfake manipulation that could misrepresent a candidate’s identity during virtual interviews. By surfacing these signals in real time, recruiters gain an additional layer of verification that helps detect potential AI fraud, ensuring hiring decisions are based on genuine candidate performance rather than AI-assisted or fraudulent behavior.
Humans are poorly equipped to detect near-perfect forgeries. Our brains evolved to spot predators, not interpolated pixels.
Deepfakes work not because they are perfect, but because our detection instincts were never trained for them.
Across all three patterns, one truth becomes obvious:
Fraud is no longer rare or exotic. It’s engineered. And hiring has become a target-rich environment.
Ophir Samson is the founder and CEO of Ezra, a voice AI interviewing platform. He combines deep technical expertise as a voice AI engineer with a decade of building teams and leading partnerships in autonomous driving at Aurora, business development at Uber, and generative AI at General Motors. He holds a PhD in Applied Mathematics from Imperial College London, was a researcher at MIT, and earned an MBA from Stanford Graduate School of Business.
