How to Tell If Someone Is Cheating on Lichess

Were they using an engine?

From a single game, you usually can't know. What you can do is understand what Lichess itself looks for, what happens when it catches someone, and how to check whether a game was statistically unusual rather than just well played.

Can you tell if someone is cheating from one Lichess game?

Not reliably.

GM David Smerdon, an economist at the University of Queensland, tested exactly this. In a controlled tournament of players rated 1600 to 2300, where some players secretly received engine moves, accusations were right about 69% of the time. In a larger online test with over 4,000 rating-verified participants, the success rate fell to around 54%, barely better than a coin flip.

An interesting finding: the strongest predictor of whether someone accused their opponent was how many blunders the accuser had made themselves.

That doesn't mean suspicion is always wrong. It means "it felt like an engine" is weak evidence on its own. A better question is: were these moves statistically unusual for a player of this rating, in positions where the right move was genuinely hard to find?

How does Lichess detect cheating?

Lichess combines automated models with human moderators. Unusually, its two best-known models are open source, so we can look at what they actually use.

Irwin: does this play look like an engine?

Irwin is Lichess's original neural-network cheat detector, the successor to an earlier system called cheatnet. It runs Stockfish over a player's games, passes the results through a TensorFlow model, and flags likely cheaters for moderators.

In the public code, every move a player makes becomes a set of features:

the player's winning chances after the move they played
how much winning chance that move gave up compared with Stockfish's best move
how long the player spent on the move, and how that compares with their average move time in the game
simple board features: how far up the board the piece moved, how many moves were available, whether it was a capture, and which piece moved

Two notable things:

Move time is a core input to the model
"How many moves were available" is a rough stand-in for how complex the position was, the same idea behind our use of position complexity to predict chess cheating.

Kaladin: does this play look like this player?

Kaladin was introduced in 2022. It uses convolutional neural networks on a player's Lichess Insights data, the aggregated statistics about how someone plays.

Lichess says Kaladin considers different information from Irwin and doesn't need a fresh engine analysis of a player's games to spot suspicious patterns. That makes it faster to run, so far more players can be screened.

Human moderators

According to Lichess, neither model has the final word and suspicious cases go to human moderators for review, apart from very obvious ones the models can handle themselves, and appeals are always reviewed by a person.

One caveat: the public repositories show how these systems were designed, not necessarily everything that runs today. Lichess has historically kept quiet about anti-cheat details, so it's safe to assume there is probably more going on than the open-source code shows.

Does Lichess look at rating or accuracy?

You'll find two opposite claims about this. Some forum answers insist rating and accuracy play no part in Lichess's detection. Some explainer articles say Lichess compares your play with what's expected at your rating. The public code can help us resolve this discrepancy.

Rating. None of Irwin's per-move inputs is the player's rating. It asks whether the pattern of moves and move times looks like engine use, not whether the play is unusually good for a given rating.

Accuracy. Irwin doesn't use the Accuracy % you see after a game, but it does use a measure of how much winning chance each move gives up. Lichess calculates Accuracy from exactly that, the change in Win% before and after each move. Kaladin works from Insights, which include accuracy statistics broken down by game phase and opening.

So accuracy-type information matters, just not as one headline number from one game. Lichess's accuracy page makes the same point: an opponent's 96% isn't necessarily superhuman, especially if you blundered early and handed them an easy game. It also notes that Stockfish can judge how sound a move is, but can not easily judge how hard it was to find.

What happens when Lichess catches a cheater?

Lichess deals with a lot of cheating. In 2025 its moderators processed 185,997 cheating reports, roughly double the 93,000 it reported for 2023. It also took 191,508 fair-play moderation actions in 2025, a figure that includes sandbagging and boosting as well as engine use.

The account is marked

Lichess's Terms of Service list a range of penalties, including a public mark on the account identifying it as having violated the Terms, account closure, and placing the account in a separate playing pool. If you suspect an old opponent, check their profile for that notice.

Opponents may get rating points back

Lichess refunds rating automatically. Per its FAQ, one minute after a player is marked, Lichess looks at their 40 most recent rated games from the last 5 days. You get a refund if:

you were their opponent in one of those games
you lost rating in that game, through a loss or a draw
your rating wasn't provisional

The refund is capped by your peak rating and how your rating moved afterwards, and never exceeds 150 points. That's why some players see a notification for only a point or two.

Cheaters can appeal

Any marked user can appeal. Lichess says its appeal panel includes experienced moderators with expertise in statistics, computer science, law and chess, including titled players rated 2400+ FIDE. It also says it never shares the reason for a mark with anyone except the account holder.

What are the signs a Lichess game might be suspicious?

No single signal proves anything. These are the ones worth looking at, roughly in order of how much they tell you.

1. Engine moves in genuinely difficult positions

Matching Stockfish when there's one obvious recapture means nothing. Matching it when there are several plausible moves and only one hard-to-see move keeps the advantage is far more informative. A long run of those is the strongest single-game signal.

2. Play far above the player's rating

The question isn't "was this a great game?" but "how likely is this game from a player at this level?" One note for Lichess: its ratings run higher than other sites and FIDE, so judge a player against the Lichess pool, not an over-the-board rating or Chess.com rating.

3. Move timing that doesn't fit the position

Humans play obvious moves quickly and slow down when a position gets hard. Warning signs include near-identical thinking time on every move, long pauses before obvious recaptures, and instant replies in positions that demand calculation.

Lichess shows the time spent on each move alongside the computer analysis, so you can check this yourself.

4. The same pattern across several games

Kaladin works from a player's accumulated statistics over time, looking for play that's out of character. You can do a simple version: open their recent games, particularly against stronger opponents. A sudden, sustained jump in quality is more telling than one brilliant game.

Weaker signals than they seem

High accuracy on its own. Short games, forced sequences and your own early mistakes all inflate it.
A new account. Plenty of honest players are new, or are returning after a break. These accounts may have badly calibrated ratings, as player strength changes over time.
"It felt like an engine." As Smerdon's research shows, that feeling tracks your own blunders more than your opponent's play. Finding strong moves may be easier when you are in a winning position.

How do you report a cheater on Lichess?

Open the player's profile and use the report icon, or go to lichess.org/report.
Choose the Cheat category, so the report reaches the right moderators.
Link the specific games and say what you noticed in each. Lichess's report FAQ is clear that vague reports are hard to act on.
Submit once, then leave it.

A few things to expect:

Moderators review every report, which Lichess says can take a day or two.
You might get a notification if the player is banned. Hearing nothing doesn't mean your report was ignored.
A second report won't help. Lichess says it neither speeds things up nor makes a ban more likely.

Don't accuse them in chat or the forums. Lichess's Terms of Service treat public cheating accusations as bad sportsmanship and a communication offence in their own right. The report form is the only place that suspicion belongs.

How can I analyze a Lichess game for cheating?

Lichess's built-in computer analysis tells you how accurate each side was and where the mistakes happened. It doesn't tell you whether that accuracy was surprising for the player, or how hard the good moves were to find.

Paste a Lichess game link into the Chess Cheat Detector analyzer, or upload the PGN, and every position is evaluated with Stockfish. The analysis then weighs:

engine agreement
centipawn loss
position complexity, so hard moves count for more than obvious ones
the player's rating, and how accurately a player at that level would be expected to play

The result is an Engine Assistance Likelihood for the game, plus the individual moves that drove it. It doesn't currently analyze clock times, so check move timing by eye using Lichess's own analysis.

Only analyze finished games. Lichess bans any outside assistance while a game is in progress, and that includes analysis tools.

Analyze a Lichess game →

Don't accuse someone based on one unusual game

Lichess's own system combines automated models, human moderators and an appeal panel, and its models assess patterns across a player's games rather than a single result.

The same standard applies when you check a game yourself. There's a difference between "this player cheated" and "this game contains statistically unusual, engine-like play." Analysis can tell you the second. Only Lichess can act on the first.

If you think someone broke the rules, report them and let the moderators decide. If you're just curious whether that game was as strange as it felt, analyze it and see what the moves say.