Aiseptor launches network-layer defense against AI exam cheating that lockdown browsers can't see

NEW YORK, NY — June 23, 2026 — Aiseptor, a network-layer exam security platform built by a Cornell Tech team, today detailed its approach to a problem the online testing industry has not solved: AI cheating tools that now operate beneath the browser and outside the webcam's view. The company won the 2026 Cornell Tech Startup Award for its work.

The cheating tools that matter in 2026 moved below the layer existing defenses watch. Invisible AI overlays such as Cluely set an operating-system flag that excludes them from screen capture, so they never appear in a screen recording or the window a lockdown browser controls. On-device language models running locally answer questions from memory with no network traffic to intercept. Lockdown browsers control a single browser window; webcam proctoring watches the candidate. Neither can see a technique designed to be invisible to both.

Aiseptor enforces integrity one layer down. A per-session encrypted tunnel routes the candidate's device traffic through a gateway that applies a default-deny policy: only assessment-approved destinations are reachable, and AI APIs, remote-access protocols, and unauthorized cloud storage are blocked before any connection completes. Detection targets the technique rather than the process name — the screen-capture-exclusion flag, GPU memory deltas, and DNS requests to AI endpoints — so renamed builds and never-before-seen overlays are caught the same way the known ones are.

“The certifications I'd spent months earning were being sold by proxy rings for $200, pay after the test clears. Every signal I'd built was equally available to someone who had never opened a terminal. The reason that works is that everyone is defending the wrong layer — the browser, the webcam, the app. We built the security layer underneath, where the cheating actually happens.”

The platform uses no webcam, no microphone, no screen recording, and no keystroke logging; its audit record is metadata about what a device could and could not reach during a session, not a recording of the candidate. Default data retention is 24 hours. The agent deploys on a candidate's own Windows or macOS device in about 30 seconds with no kernel driver and no persistent install, and removes itself after the session ends. Aiseptor is designed to sit beneath an existing lockdown browser or proctoring service — covering the device and network surface they cannot reach — rather than replace identity or physical proctoring.

In a live pilot with more than 100 students at Cornell University, Aiseptor recorded a 0% bypass rate and zero false positives across the attack vectors attempted. The company has signed three paid pilots with assessment platforms and has interviewed more than 100 assessment and talent leaders while building its enterprise product.

Aiseptor's approach is patent pending. The company was founded in 2025 by Akshay Aggarwal, Divya Bhanushali, and Sanjay Ram. Qualified institutions and teams can request a limited number of free sessions at aiseptor.com/try-free.