Can Sensei Copilot Be Detected? The macOS Interview Assistant

By Vaibhav Devere, Founder, Zero Assist · 2025-05-19 · 5 min read

What Is Sensei Copilot

Sensei Copilot is a macOS application that provides an AI assistant overlay on top of any running application. Unlike browser-based tools, it is a native macOS app that:

  • Floats above all other windows with an "always on top" flag
  • Captures screen content using macOS screen recording permissions
  • Sends screenshots to an LLM for analysis and answer generation
  • Displays answers in a compact overlay that can be resized and repositioned
  • Supports keyboard shortcuts for quick hiding and showing

Its native macOS architecture makes it particularly difficult to detect with browser-focused or Windows-centric proctoring tools.

How Candidates Use Sensei Copilot

During Screen Sharing

Even when screen sharing is active:

  • The overlay window can be positioned in a corner the interviewer is unlikely to notice
  • Some video conferencing tools do not capture "always on top" windows in screen shares
  • The overlay can be set to a very low opacity, making it nearly invisible on camera
  • Keyboard shortcuts allow instant hiding if the candidate suspects detection

In IDE-Based Interviews

For live coding interviews:

  • The candidate takes a screenshot of the problem statement using a hotkey
  • Sensei Copilot sends the screenshot to its backend
  • The answer appears in the overlay within seconds
  • The candidate types the solution while the overlay remains visible

For System Design Questions

  • The candidate describes the system verbally
  • Sensei Copilot listens (if microphone access is granted) or the candidate types a summary
  • The tool generates architecture diagrams and component explanations
  • These are displayed in the overlay for the candidate to read from

Can Sensei Copilot Be Detected

Yes — through macOS process monitoring, window enumeration, and screen capture permission auditing.

Process Detection

Sensei Copilot runs as a standard macOS application:

  • Bundle identifier and executable name in the process list
  • Helper processes for screenshot capture and LLM communication
  • Menu bar icon (even if hidden, the process remains)
  • LaunchAgent or LaunchDaemon for auto-start behavior

MacSystemScanner using macOS APIs can enumerate:

  • All running applications via NSWorkspace
  • Window lists via CGWindowListCopyWindowInfo
  • Process trees via ps and lsof
  • Recent items and download history

Window Enumeration

The overlay window is detectable through macOS window APIs:

  • CGWindowListCopyWindowInfo returns all windows including "always on top" overlays
  • Window properties include bounds, opacity, owner process, and layer level
  • A window that is simultaneously: always on top, owned by a non-system process, and partially transparent is a strong indicator

Screen Recording Permissions

Sensei Copilot requires screen recording permissions to function:

  • macOS Privacy & Security settings log which apps have screen recording permission
  • A newly granted permission for an unknown app before an interview is a red flag
  • The permission can be checked programmatically via the ScreenCaptureKit API

Network Traffic

When analyzing screenshots and generating answers:

  • Uploads to the Sensei Copilot backend servers
  • LLM API calls (may use OpenAI, Anthropic, or custom models)
  • Image analysis API calls for screenshot interpretation
  • WebSocket connections for real-time streaming

Behavioral Signals

Even with the overlay hidden, candidates exhibit tells:

  • Reading from a source not on the shared screen
  • Eyes tracking text that is not visible to the interviewer
  • Sudden improvements in answer quality when the overlay is activated
  • Typing patterns that match reading and copying rather than thinking and composing

The macOS Ecosystem Challenge

macOS has a rich ecosystem of AI overlay and assistant tools:

  • Sensei Copilot, Parakeet AI, and similar tools exploit macOS window management
  • ScreenCaptureKit and AVFoundation provide powerful APIs for capture
  • The App Store and direct downloads make these tools easily accessible
  • Many candidates interviewing for Silicon Valley roles use macOS as their primary machine

A proctoring solution that only supports Windows misses a significant portion of the threat landscape.

Bottom Line

Sensei Copilot is detectable on macOS through native APIs that enumerate processes, windows, and permissions. However, its effectiveness as a cheating tool highlights the need for cross-platform monitoring. Organizations that only deploy Windows-compatible proctoring are leaving their macOS-using candidates unmonitored.

OS-agnostic, forensic-level monitoring is essential for comprehensive interview integrity.