Can LockedIn AI Be Detected? The Interview Automation Bot

By Vaibhav Devere, Founder, Zero Assist · 2025-05-21 · 6 min read

What Is LockedIn AI

LockedIn AI is an interview automation tool designed to conduct technical interviews on behalf of candidates. Unlike assistance tools that provide hints or code suggestions, LockedIn AI aims to fully automate the interview process:

  • Listens to the interviewer's questions via audio capture
  • Transcribes the question using speech-to-text
  • Generates a complete answer using a large language model
  • Converts the answer to speech and plays it back to the interviewer
  • Optionally types code into the IDE autonomously

The tool represents a significant escalation in interview cheating technology — moving from assistance to full substitution.

How LockedIn AI Works

Audio Pipeline

The core of LockedIn AI is its audio processing pipeline:

  1. Captures audio from the candidate's microphone input
  2. Uses a secondary audio channel or software loopback to receive the interviewer's voice
  3. Applies noise cancellation and voice isolation
  4. Feeds the transcribed question to an LLM

Answer Generation

Once the question is transcribed:

  • The system selects the appropriate model based on question type
  • Generates a spoken response and/or code solution
  • Applies voice cloning to make the output sound like the candidate
  • Introduces realistic pauses and filler words to sound natural

Code Automation

For coding portions:

  • The tool can interface with common IDEs through accessibility APIs
  • Types code character by character with human-like typing speed
  • Introduces intentional typos and corrections to appear authentic
  • Can run and debug code autonomously

Can LockedIn AI Be Detected

Yes — though it requires sophisticated multi-layer detection because the tool is designed to mimic human behavior.

Audio Analysis

The audio pipeline creates detectable artifacts:

  • Duplicate audio streams: The tool must capture both input and output audio, creating unusual audio routing configurations
  • Loopback devices: Software audio loopbacks (like VB-Cable or BlackHole) appear as additional audio devices in the system
  • Voice synthesis signatures: Even advanced voice cloning has subtle artifacts — unnatural breath patterns, consistent pronunciation of ambiguous words, lack of emotional variation
  • Latency patterns: The delay between question and answer is longer than human reaction time but shorter than genuine thinking

System-Level Detection

LockedIn AI requires significant system resources and permissions:

  • Accessibility API access for IDE automation (screen reader permissions)
  • Microphone exclusivity or audio routing software
  • Background processes for speech-to-text and text-to-speech
  • Network connections to LLM APIs during the interview

Forensic monitoring can detect:

  • Processes with names containing speech recognition libraries (Whisper, Vosk, etc.)
  • Accessibility service registrations that appeared just before the interview
  • Audio routing configurations with virtual devices
  • Network traffic to LLM APIs with interview-timed patterns

Behavioral Red Flags

Despite its sophistication, LockedIn AI has behavioral tells:

  • Eye line: The candidate is not looking at the screen during technical explanations because they are not the one thinking
  • Typing mismatches: The IDE shows typing activity while the candidate's hands are visible away from the keyboard
  • Answer perfection: Responses that are too polished, too complete, and never include the candidate's personal experience or uncertainty
  • No self-correction: Humans hesitate, backtrack, and reformulate. Automated answers flow perfectly from start to finish.

The Ethics and Legal Dimension

LockedIn AI blurs the line between cheating and fraud. Using assistance tools is an integrity violation; using a substitute to conduct an interview on your behalf may constitute:

  • Fraudulent misrepresentation
  • Breach of employment contract terms
  • Identity misrepresentation

Organizations should explicitly state in their interview policies that AI automation tools, not just assistance tools, are prohibited.

Bottom Line

LockedIn AI is detectable, but not by traditional proctoring. Its reliance on audio routing, accessibility APIs, and voice synthesis creates multiple detection vectors. However, the most reliable detection remains behavioral: a candidate who never hesitates, never backtracks, and never shows genuine uncertainty is either a savant or using automation.

Structured follow-up questions that probe understanding, not just output, will expose automated responses faster than any technical monitoring.