Candidate Cheating Techniques with AI and How to Stop Them

TL;DR

Candidates now use AI during interviews to:

• Generate answers live

• Receive real-time coaching

• Deepfake their face or voice

• Copy solutions into code editors

• Have another person answer in the background

The strongest defense is not asking harder questions.
It is designing your interview workflow to detect signal mismatches in reasoning, behavior, and response latency.

Sherlock AI provides automated detection.
We also list manual countermeasures when tools are not in place.

The Five Most Common AI-Enabled Cheating Techniques

1. Live Answer Whispering Through a Second Device

Candidate uses:

• Phone just outside camera view

• Earbuds to receive instructions

• Co-pilot software providing real-time hints

Detection Signals:

• Answers sound correct but lack detail

• Delay between question and response is constant

• Candidate never asks clarifying questions

How Sherlock AI Stops It:

• Audio cross-channel analysis

• Background conversation fingerprinting

• Earbud and silence-pattern detection

Manual Countermeasure (if no tool):

Ask:

Before we continue, explain the problem in your own words.

If the paraphrase is weak, the answers are not their own.

2. Reading From a Script or Prompt Sheet

Often used in:

• System design interviews

• Project walkthroughs

• Behavioral interviews

Scripts are often purchased from coaching institutions.

Detection Signals:

• Highly polished narrative

• No mention of mistakes or course corrections

• Story structure identical to industry templates

How Sherlock AI Stops It:

• Narrative pattern matching

• Reasoning adaptability checks under constraint shift

Manual Countermeasure:

Ask:

What changed during the project after launch?

Real experience always contains change.
Scripted answers nearly never account for it.

3. AI-Generated Coding Solutions

Candidate:

• Pastes solutions from ChatGPT or GitHub Copilot

• Cannot debug or modify under constraint change

More than 50% of candidates use AI tools in virtual coding interviews.

Detection Signals:

• Code is correct but style shifts mid-answer

• Variable naming patterns inconsistent

• Candidate struggles when asked to optimize or restructure

How Sherlock AI Stops It:

• Typing cadence analysis

• Copy-paste detection

• Code lineage pattern recognition

Manual Countermeasure:

Ask:

Walk me through how you would improve performance if input size increased by 10 times.

Authentic coders reason forward.
Cheaters freeze.

4. Proxy Interviewer (Another Person Answers)

Most common in:

• Remote engineering interviews

• Contracting roles

• Global sourcing pipelines

Detection Signals:

• Candidate avoids turning on camera

• Explanation style and voice confidence mismatch resume claims

• Real knowledge only appears when improv-level detail requests are avoided

How Sherlock AI Stops It:

• Face match verification

• Voice identity comparison

• Interaction pattern detection over time

Manual Countermeasure:

Ask:

Could you open your settings and switch cameras?

Proxy setups usually collapse at this request.

5. Deepfake or Masking Software During Video Interviews

Still emerging, but increasing rapidly.

Detection Signals:

• Blurry skin texture only around face

• Mouth movement does not align perfectly with speech

• Eye gaze does not react to screen content

How Sherlock AI Stops It:

• Liveness checks

• Gaze tracking alignment

• Head movement realism signals

Manual Countermeasure:

Ask candidate to:

Turn head slightly to the left and say the phrase: "Testing for clarity."

Deepfakes struggle with side profile rendering.

Why Trying to “Outsmart” Candidates Doesn’t Work

Raising difficulty leads to:

• Anxiety for honest candidates

• Advantage for coached candidates

• Interviewer inconsistency

The correct approach is:

• Standardized prompts

• Consistent reasoning checkpoints

• Automated authenticity signals

The Preventive Interview Workflow (Copy This)

1. Verify identity before the interview starts

2. Use one micro problem to test reasoning

3. Apply one constraint shift to verify adaptability

4. Record findings directly in the scorecard

5. Escalate only when reasoning and behavior do not match resume claims

This workflow is:

• Light

• Fair

• Hard to cheat

• Easy to train globally

Conclusion

AI cheating is not a fringe phenomenon.
It is a predictable pattern with detectable signatures.

The goal is not to catch criminals.
The goal is to ensure the person hired is the same person who performs the job.

With:

• Real-time AI-based fraud detection (Sherlock)

• Reasoning-focused interview prompts

• Consistent documentation standards

Companies maintain:

• Talent quality

• Fair candidate treatment

• Audit safety

• Brand integrity