What it is
A transparent heads-up display that streams AI-generated answers directly to a candidate's local screen while remaining invisible to screen-sharing, webcam, and most proctoring software.
Why it matters
Invisible overlays are the dominant AI cheating vector in 2025–2026, and they defeat lockdown browsers and behavioral proctoring by design; nothing about the captured video stream reveals their presence.
How Aiseptor addresses it
Aiseptor operates below the application layer, so overlay tools and their traffic patterns cannot hide from its enforcement. The exam device is constrained to the authorized network path regardless of what renders on the candidate's screen.
Definition
Canonical definition
An invisible AI overlay is a desktop application that uses low-level graphics hooks to render text on top of other windows while marking its own surface as excluded from screen-capture APIs. The candidate sees AI-generated answers, code, or prompts in a heads-up display aligned to the interview question. A remote interviewer, proctor, or screen-share viewer sees only the underlying editor; the overlay is absent from the captured video stream. Audio variants extend this pattern: the candidate's microphone is piped to a local speech-to-text pipeline that feeds a language model, and the answer is shown on the same invisible surface. Because the overlay produces no artifacts in a standard screen recording, detection strategies built around what the interviewer sees are architecturally defeated. The category includes both commercial products and the many open-source forks that ship with the same primitives. The exclude-from-capture flag these tools rely on is not a hidden exploit; it is a legitimate OS feature originally intended for things like password managers avoiding accidental exposure during screen-shares, repurposed for the opposite goal of hiding assistance rather than protecting secrets.
In practice
A concrete example
An interviewer runs a standard video call with screen-share enabled and watches the candidate's editor the entire time. The candidate has an overlay running that captures the spoken interview question via microphone, sends it to a cloud model, and renders the answer as a small floating panel positioned just above the code editor. On the interviewer's screen, none of this exists: no panel, no flicker, no visual artifact. The candidate appears to type a working solution unprompted, because from the interviewer's side of the screen-share, that is exactly what is happening.
Citations
- [1]Fabric, analysis of 19,368 AI-conducted interviews, January 2026 (2026)
- [2]Talview AI Threat Index Report 2026 (2026)
- [3]Aiseptor threat intelligence log on overlay tooling (2026)