Not all anti-cheat happens live. Some of the most convincing evidence a cheater leaves is the match itself — and replay review is how games make use of it.
Why the recording is evidence
Many games record matches as replays or demos that can be played back later. A replay captures what every player did: every movement, every shot, every decision. For catching cheats, that is valuable, because cheating shows in behaviour. Aim that snaps with impossible precision, awareness of enemies through walls, reactions faster than human reflexes — these are visible when you watch the recording back, even if the cheat software itself left no trace.
Human review
The simplest use is people watching. Reviewers — sometimes dedicated staff, sometimes trusted community members in structured systems — watch replays of reported or flagged players and judge whether the behaviour is humanly possible. It is slower than automated detection, but a careful human watching a full match is very hard to fool, because they are judging the result, not searching for code.
Feeding machine detection
Replay review also feeds automated systems. When humans label a large set of replays as cheating or legitimate, that labelled data can train machine-learning models to recognise the same patterns at scale. A well-known anti-cheat for a major shooter was trained in exactly this way, on data from a community replay-review process. Human review and machine detection reinforce each other.
Why it is hard to beat
Replay review is powerful because it sidesteps the whole cheat-versus-anti-cheat technical race. A cheat can be rewritten to hide from scans, but it cannot rewrite the match that was already recorded. The behaviour is the evidence, and the behaviour is what the cheat exists to produce. To beat replay review, a cheater would have to play like a human — at which point they are not really cheating.
The takeaway
Replay review catches cheaters by examining the match itself rather than scanning the PC. It works through human reviewers and through machine-learning models trained on their judgements. Its strength is that it judges behaviour, which is the one thing a cheat cannot hide — the match already happened.