How BlackGuard's AI actually works

Every anti-cheat says "AI" now. Almost none explain what that means. Here's the actual mechanism behind BlackGuard's detection: what runs where, what the AI layers actually judge, and — the part that matters most to your players — what gets auto-banned versus what gets flagged for your team to decide.

It runs server-side — nothing on the player's PC

BlackGuard's detection runs server-side. There's no client-side agent, no kernel-level driver, nothing installed on the player's machine for a cheat developer to pull apart. That matters for two reasons: there's no local file a cheat can patch to blind the anti-cheat, and there's no client binary for someone to reverse-engineer and build a targeted bypass against. The server sees what actually happened in the game world — it doesn't have to trust anything the player's PC reports about itself.

Behavioral AI and Vision AI: judging behavior, not names

Signature-based detection matches a known cheat's name or file fingerprint — and falls behind the moment a cheat gets renamed or repacked. BlackGuard's two AI layers don't work that way; they judge what a player does and what's actually on their screen, so they don't need to have seen the exact cheat before.

Behavioral AI looks at gameplay-data patterns — aim behavior, movement — and an AI model judges whether what it's seeing looks like normal human play or something outside that range. Vision AI takes secure screenshots and analyzes the rendered image itself for visual cheat indicators: aim overlays, ESP boxes, wallhack visuals — the things a cheat draws on screen that behavioral data alone would never show. Vision is live today, available on the Pro tier and up; Starter runs the server-side deterministic engine plus the behavioral AI layer, without Vision. Vision currently detects aimbot and aim assist, ESP, no-clip, and god-mode; captured images auto-delete after 14 days.

What gets auto-banned versus what gets flagged

This is the part most "AI anti-cheat" marketing skips, and it's the part that actually protects your players. BlackGuard auto-bans only physically-impossible events — speed, teleport, and impossible weapon damage — measured deterministically on the server. There's no AI judgment call in that path at all: if the movement or damage was physically impossible, it's caught and actioned immediately.

Everything the AI layers surface — behavioral suspicion, resource injection, economy and money-exploit patterns (on ESX and QBCore), and Vision matches — is flagged for your own staff to review and action. None of it is auto-banned by AI. Probabilistic signals never auto-act, because banning a real player on a guess is worse than a cheater staying online a little longer. See the full breakdown of why BlackGuard is built to avoid false bans.

Why this design, not the alternative

An anti-cheat that lets an AI model auto-ban on a hunch will eventually ban a legitimate player having a good day. An anti-cheat that only reacts to the physically-impossible misses the cheats that don't break the laws of the game world — aim assist, ESP, injected resources. BlackGuard's answer is to run both: a deterministic engine that acts instantly and without ambiguity, and AI layers that extend visibility into what a rule-based engine alone can't see, without ever being handed the ban button themselves. For the fuller architecture picture, see how this compares to other FiveM anti-cheats.

FAQ