Proctoring Software for Assessment Platforms: The Shift to Network-Layer Integrity in 2026

Proctored coding-assessment fraud more than doubled in a single year, from 16% to 35% between 2024 and 2025, according to CodeSignal. Separately, Talview's 2026 AI Threat Index puts 88% of online assessments at active AI cheating risk. If you run an assessment platform, both numbers describe the same underlying problem: the proctoring layer your customers already trust wasn't built to see where this cheating actually happens.

Here's why browser- and camera-based proctoring is structurally blind to modern AI cheating, and what an assessment platform needs from a security layer to close that gap without rebuilding its own delivery engine.

Why Browser-Based Proctoring Is Blind to Modern AI Cheating

Lockdown browsers and webcam proctoring were built against an older threat model: a candidate opening a second tab, or looking up an answer on their phone for an extended stretch. Against that threat model, they still work. The tools candidates use now don't touch the browser at all.

Invisible Overlays

An invisible AI overlay like Cluely renders on top of the exam window while marking itself excluded from screen-capture APIs. It never enters the video stream a proctor reviews, and it never touches the browser's DOM, so a tool built to watch either surface has nothing to flag.

On-Device LLMs

A language model running locally on the candidate's machine (via Ollama, LM Studio, or a custom runtime) generates no request to a known AI provider's API. Detection requires OS-level signals, active inference processes, GPU memory deltas, model files on disk, none of which a browser sandbox or a webcam has any way to observe.

What Network-Layer Integrity Looks Like

Aiseptor enforces a per-session, default-deny network policy on the candidate device: only the assessment platform and any resources it explicitly allows are reachable for the duration of the session. AI inference endpoints, remote-access relays, and model-hosting services are unreachable, regardless of what the requesting tool is named or how it disguises itself. This replaces a name-based denylist, which is defeated the moment a tool is renamed or recompiled, with an approach that targets the underlying technique instead.

The enclave is ephemeral: it deploys on the candidate's device in about 30 seconds, requires no kernel driver, and removes itself completely when the session ends. No webcam, no microphone, no keystroke logging. Aiseptor is GDPR- and CCPA-compliant by design, with a default 24-hour data retention window, and is currently undergoing a SOC 2 Type II audit targeted for completion in Q3 2026.

Integrating Into an Existing Platform

Assessment platforms don't need a standalone application; they need a security layer that drops into a session lifecycle they already own. Aiseptor's REST API triggers an enclave at the start of a session and reports a signed integrity verdict at the end, typically integrated in under a business day. This is deliberately narrow in scope: it doesn't replace your delivery engine, your scoring logic, or your existing identity checks. It closes the one layer those systems were never built to reach.

Pricing is usage-based, billed per session, with no seat-based minimums, which matters for platforms whose session volume moves with customer demand rather than a flat annual number.

Where This Fits Alongside Existing Tools

Network-layer enforcement is a complement to identity verification and behavioral proctoring, not a replacement for them. Those tools remain the right layer for confirming who is sitting the exam and catching physical-room anomalies. Aiseptor covers the device and network surface underneath: the layer where invisible overlays, on-device LLMs, and remote-access tools actually operate. See Aiseptor for assessment platforms, or read the full architecture.

Frequently Asked Questions

How does network-layer proctoring differ from a standard lockdown browser?

A lockdown browser restricts a single application window. Network-layer proctoring controls what the candidate device can reach on the network for the duration of the session, which covers invisible overlays and on-device LLMs that never touch the browser at all.

Can this detect tools like Cluely?

Aiseptor targets the technique overlays like Cluely must use to stay hidden from screen capture, plus the network endpoints they depend on, rather than matching a specific process name.

Does the candidate need to install permanent software?

No. The enclave is session-scoped and removes itself completely once the exam ends.

How long does integration take?

Typically under a business day via REST API, with no changes required to your existing delivery engine or identity checks.

How is it priced?

Usage-based, billed per session, with no long-term seat commitment required.

Is this GDPR compliant?

Yes. Aiseptor collects network-access signals and device-activity metadata only, no webcam, microphone, or keystroke data, with a default 24-hour retention window.