Any phone in any pocket contains a chess engine that beats every human who has ever lived. That single fact defines online chess: the question was never whether cheating is possible, but whether it can be detected reliably enough that honest players keep playing.
Here's how detection actually works — described at the level any player can understand, and deliberately without the thresholds that would help someone game it.
The naive approach is to compare a player's moves to an engine's top choice and flag anyone above some match percentage. It fails in both directions, badly.
False positives: in simple positions, strong humans and engines agree constantly. A forced recapture sequence is 100% "engine match" and proves nothing. Endgames with few pieces, tactical sequences with one legal answer, well-known opening theory — all produce high match rates from honest play.
False negatives: a careful cheater doesn't play the top move every time. They consult the engine in three critical positions per game and play their own moves elsewhere. Match percentage barely moves; the games are still stolen.
So every serious system layers multiple independent signals instead.
1. Move quality relative to the player. The useful question isn't "how good were these moves?" but "how good were these moves for this player?" A 1400-rated player finding a deep queen sacrifice with a quiet follow-up is far more suspicious than a 2400 finding it. Modelling what a human at a given strength plausibly plays — rather than what a maximally strong engine plays — is the core of modern detection.
2. Timing. Human thinking time correlates with position complexity: hard positions take longer. Engine-assisted play flattens that curve, and sometimes inverts it — instant replies in complex positions, or a suspiciously constant delay while a player reads a screen and copies a move.
3. Behavioural and interaction signals. How a player interacts with the board — patterns of input, focus, and the rhythm of play — differs measurably when moves are being transcribed from elsewhere rather than chosen.
4. Account-level and stake-level patterns. Cheating for money has a shape: which opponents, at which stakes, with which results, from which devices and networks, and how a new account's results relate to its claimed strength. Looking at one game in isolation misses all of it.
5. Human review. Every automated system produces ambiguous cases, and on a real-money platform an ambiguous case is somebody's money. The final call on serious verdicts should be a person looking at the evidence, not a threshold firing silently.
Each signal on its own generates false positives. A genuinely brilliant game from an improving player will look unusual on move quality. A player on a slow connection or a phone will look unusual on timing. A new account from a shared network will look unusual at the account level.
Combining independent signals is what makes detection safe: the honest explanations rarely line up across all of them simultaneously, while assistance tends to leave traces in several at once. ChessBit's system, Argus, is built on that principle — four layers that have to agree — and it's described in more depth in inside Argus.
On a free site, catching a cheater means correcting a rating. On a real-money site, it means someone's stake. That changes the design in two ways:
- Payouts clear before they land. Winnings sit in escrow while the game is checked, rather than being paid instantly and clawed back later.
- The honest player is made whole. On ChessBit, a game voided for cheating returns the honest player's full stake plus a bonus equal to the fee the cheater paid — and we refund our own fee too, so the platform earns nothing on a cheated game. A platform that profits from cheated games has the wrong incentive; a platform that loses money on them has the right one.
- Play on platforms that publish how they handle it. Vagueness about anti-cheat is itself a signal.
- Report suspicion rather than accusing publicly. Public accusations are usually wrong and always corrosive; use the reporting channel, where the evidence can be checked properly.
- Know that fair-play holds are normal. A short delay on a payout while a game clears isn't an accusation — it's the system working for you as much as against anyone.
ChessBit is a skill-based real-money chess platform in early access, 18+. Fair-play policy: Argus FAQ.
