Live Interview Verification & Remote Interview Fraud Detection

Signal coverage

Checks that screen share alone cannot provide.

Fourteen detectors watch the interview environment across two loops. Each maps a common cheating method to objective, machine-level evidence.

AI & knowledge assistance

AI desktop apps, browser AI tabs, local LLM runners, extension assistants, and live visits to knowledge sources like Stack Overflow or docs.

Hidden & external help

Capture-excluded overlays, remote-control and proxy tools, and focus switches to unexpected apps.

Environment & timing

Monitors added or removed mid-interview, virtual camera and audio drivers in use, clipboard bursts, and pause-then-burst answer timing.

How it works

A two-speed engine that turns events into evidence.

A fast loop samples window, title, focus and clipboard every two seconds; a slow loop sweeps processes, network/DNS and devices every five. Detector output collapses into incidents, and a heartbeat every five seconds means a silent agent is itself a signal.

Durative signals get start/update/end with a duration; momentary ones are point events. Every envelope is signed and sequenced.

Live timeline

Every signal has timing and context.

InterviewWatch never reduces a session to a vague suspicion. The report shows when a signal appeared, how serious it was, and what happened around it.

14:32

AI assistant detectedBrowser AI tab became active during the architecture question.

High

14:36

Clipboard burstLarge structured paste occurred after a long reading pause.

Review

14:41

Overlay checkNo screen-share-excluded overlay was present.

Clear

14:55

Heartbeat continuousAgent reported throughout, no gaps or tamper signals.

Intact

Why correlation matters

One event is noise. Aligned events are evidence.

Signal correlation, not gut feel.

A clipboard paste on its own is harmless. A clipboard paste that lands seconds after an AI tab opens, during the hardest question, next to a low-variance typing burst, that is a pattern. InterviewWatch scores the combination, so reviewers see the story rather than a pile of disconnected alerts.

2s / 5sFast and slow sampling loops running in parallel

14Independent detectors mapped to real cheating methods

0Pixels of screen, audio or video recorded

Integrity metadata only

InterviewWatch observes process, window, device, focus, clipboard-size and timing metadata relevant to interview integrity. Nothing more.

No private media, no auto-reject

No screen, audio or video recording. No eye or body tracking. No keystroke content. And no automatic rejection, findings support a human decision.

FAQ

Live verification questions, answered.

Does InterviewWatch record the interview?

No. It records structured integrity signals only, process, window, focus, device, clipboard size and timing metadata. There is no screen, audio, video or keystroke-content recording.

Which video platforms does it work with?

It is platform-independent. Verification runs at the operating-system level, so it works alongside Zoom, Microsoft Teams, Google Meet or any other call. The meeting platform itself is only noted as context and never deducts points.

How does it avoid false positives?

Every signal carries timing, severity and context, and the score combines independent signals rather than reacting to one event. A single clipboard paste is minor; a paste right after an AI tool appears is treated very differently. Nothing is auto-rejected.

What happens if the candidate closes the agent?

The agent emits a heartbeat every five seconds. A missing heartbeat is itself a tamper signal, and events are signed and sequenced so gaps and reconnects show up in the report.

Verify the environment, not just the video.

Give every remote interview an objective, tamper-evident integrity record, without recording a single pixel.

Try now See the full platform