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 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.

“Certifications are being sold by proxy rings for $200, pay after the test clears, and every signal a legitimate candidate builds is equally available to someone who has never opened a terminal. That works because everyone is defending the wrong layer: the browser, the webcam, the app. We built the security layer underneath, where the cheating actually happens,” says Akshay Aggarwal, describing why the team built Aiseptor this way.

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.

The company says an early pilot at Cornell University, run across more than 100 students, recorded a 0% bypass rate and zero false positives across the attack vectors attempted, and that it has since signed paid pilots with several assessment platforms while interviewing more than 100 assessment and talent leaders in building its enterprise product. Aiseptor has not yet published independent third-party verification of these results.

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.