Proctoring Software That Detects AI: What to Look For

The architectural question that decides everything

A tool's detection ceiling is set by where it runs. Extensions and lockdown browsers observe the browser; a desktop agent observes the machine. Everything else in an evaluation is downstream of that single fact.

This is not an argument that browser-based tools are bad. They handle tab control, navigation restriction, and identity checks well. It is an argument that they were never designed to see a native application drawing over the exam, and no configuration changes that.

Why browser-based tools cannot see the assistance

Browser extensions and lockdown browsers observe the page, the tab, and often the webcam. Their visibility ends where the browser process ends, and that is precisely where modern AI assistance lives.

Desktop assistants and answer overlays render above every other window, are frequently excluded from screen capture and screen sharing, and never interact with the assessment tab. Remote-control tools, virtual machines, injected keyboard input, and a second device off camera are equally invisible from inside the browser.

• Native overlay applications draw over the assessment and are often excluded from screen share.
• Remote-access tools allow another person to drive the machine with no browser-visible trace.
• Virtual machines let the assessment run inside a controlled environment with helpers outside it.
• Injected or synthesized keyboard input can enter answers that were never physically typed.
• A second phone, tablet, or monitor off camera leaves nothing at all in the browser.

What device-level detection observes

Reliable detection has to happen at the layer where assistance runs: the operating system. ScreenComply pairs a browser-based API with an optional desktop agent that inspects the environment around the assessment rather than only the page inside it.

The agent enumerates running processes and browser extensions against a maintained list of AI assistants and answer overlays, looks for always-on-top or transparent windows, detects remote access and virtualization, enumerates connected displays from the operating system, verifies that keyboard input originates from local physical hardware, and records focus changes, clipboard events, and typing-cadence patterns.

No detection system catches everything, and ScreenComply does not claim otherwise. The objective is corroborating, timestamped evidence that lets a human reviewer make an informed and defensible decision.

• Process and extension enumeration against a maintained AI-assistant list.
• Overlay, transparency, and always-on-top window detection.
• Remote access, virtual machine, and screen-capture driver detection.
• Display enumeration reported by the operating system, not inferred from gaze alone.
• Input-source integrity, clipboard events, and typing-cadence analysis.

Questions worth asking every vendor

Most evaluation checklists compare feature names. These questions compare what a tool can actually observe and what it hands a reviewer afterwards.

• Does the software run in the browser, on the device, or both — and what does each mode see?
• Which signal families are collected: process, overlay, input source, clipboard, display enumeration, remote access, virtualization, network?
• How is the list of AI assistants and overlay tools maintained as new ones appear?
• Can it block restricted applications, or only flag them after the fact?
• What does a flagged session produce: a score, or a timestamped evidence timeline a reviewer can defend?
• What retention modes exist, and can artifacts be minimized or not stored at all?
• How are assistive technologies and accommodations handled?
• Does it run alongside the LMS, ATS, or assessment platform already in place?

Claims that should make you slow down

Be cautious with any vendor promising complete detection, a single trustworthy AI-probability score for written text, or certification status that is asserted rather than evidenced. For the record: ScreenComply's SOC 2 Type 2 examination is in progress, and the company is under contract with the State of Montana.

The honest framing is that detection raises the cost and lowers the success rate of assistance while producing evidence for review. Any stronger claim than that deserves scrutiny.

Turning signals into a report someone can defend

A detection is only useful if it can be explained to a person who was not in the room. Every session produces a structured integrity report: an executive verdict with a graded risk level, a chronological evidence timeline, and the underlying signal detail behind each entry.

Language is deliberately factual rather than accusatory, because the institution or employer makes the decision, not the software. Reports export to PDF and can be shared through a secure link for academic-integrity hearings, HR reviews, and compliance audits, with configurable retention including zero-retention operating modes.