How MLSAC detects cheaters with neural networks

Step 1: the plugin watches combat

During a fight the MLSAC plugin records, tick by tick, how a player turns the camera, where the crosshair is relative to the target, when and how they click, whether hits land, how they move and whether a block stands between them and the target. Lag compensation makes sure the plugin sees the target where the player actually saw it.

Step 2: neural networks score short windows

The fight is cut into short windows of a few seconds, and each window is sent to the MLSAC cloud. A neural network trained on many hours of real legit and cheating gameplay returns a probability that the window comes from a cheater.

One window is never enough to punish anyone. The score feeds a detection buffer that grows on confident windows and slowly decays on clean play.

Step 3: buffer, flags and what happens next

When the buffer fills up, the player gets a flag. The server owner decides on the website what each flag means: an alert to staff, a kick, a ban, or suppression — reduced damage, extra damage taken, lag or inventory tricks. Suppression can also start earlier, when the buffer passes chosen thresholds.

Why machine learning

Modern combat cheats are tuned to stay inside fixed limits: they hit at normal reach, add random misses and smooth their aim. A model that has seen how real players move their mouse and click can still tell the difference, and it keeps improving as new recordings of cheats and legit players are added.