Server-authoritative anti-cheat that plays the long game.
Guardian.AI watches player behavior over time instead of punishing the instant something looks unusual — so real exploitation gets caught, and lag spikes don't get anyone banned.
All detection and enforcement logic runs inside ServerScriptService — no client-side Executor can ever read, edit, or delete it.
Live session, watched server-side
ALL CLEARModular, server-owned architecture.
Detection modules are self-contained pieces that plug into a shared engine, so new protections can be added, removed, or tuned without reworking the whole system.
Detection modules operate independently and report findings to a central engine — the only part of the system that holds shared state. Nothing about that engine is reachable from the client.
All modules and the engine live server-side. The client never touches this layer.
Three modules, watching different things.
Each one covers a different way exploiters try to get in — and each reports independently to the central engine.
Movement Detector
Validates position and velocity server-side to catch teleport hacks, speed hacks, and jump-power (JP) hacks — the checks a client-side script can never see, let alone spoof.
Remote Guard
Rate-limits and validates every RemoteEvent call, telling real exploitation apart from legitimate rapid input so fast, honest players are never mistaken for spam.
Honeypot Triggers
Hidden server elements no real player can ever interact with. If anything fires one, it's an automated exploit script — not a false positive.
Two meters decide what happens next.
Heat absorbs ambiguous signals like lag spikes and desync — it only changes how closely a player is watched, never triggers punishment. Suspicion is fed only by actions a normal connection could never produce, and it's what actually drives enforcement.
| Tier | Trigger | Action |
|---|---|---|
| All Clear heat-driven |
Normal play | No action — standard monitoring |
| On the Radar heat-driven |
Heat crosses threshold | Passive only — increased scrutiny and logging |
| Quarantined suspicion-driven |
Suspicion reaches 100% | Temporary ban, time-boxed |
| Sanctioned suspicion-driven |
Certain-confidence offense, or 2nd Quarantine | Permanent ban, logged and alerted |
Three stages, each a complete product.
Guardian.AI isn't one all-at-once launch — every stage ships as something a game can fully run on its own.
The Automated Core
A self-contained, local Roblox package. Automated movement tracking, honeypot remotes, and the full Heat & Suspicion engine — distributed and updated through Roblox Packages.
Team Integration
Adds moderator commands, config-driven permission roles, and a full action audit log. Introduces global logging with an optional in-game panel and one-way Discord alerts.
The Web Ecosystem
Guardian.AI's own backend and web dashboard. Core detection logic moves to the cloud, so updates roll out instantly to every connected game — no republishing required.
The goal: "Secured by Guardian.AI" on every top studio's loading screen.
Guardian.AI is in active development. Reach out for early access to the current build and setup docs.