I bet I would play already 200 ELO better if I'd only check if my move isn't a complete tactical blunder. This obviously begs the question if engine cheating can be as easily detected if

  • I'd only use an engine that way
  • I'd constantly play only the second- or third-best move (of course there are forced moves, but it is probably unsuspicious to recapture the queen just traded...)
  • I only test critical decisions

After all, detection methods are meanwhile far more advanced than "plays the engine move". (I once managed to play the tablebase move for 10 or so moves in an endgame, but then, I'm very good at endgames...)
By the way, the questions would also be interesting for normal computer chess.

Disclaimer: You can tell me, as I'll never played online chess in my life and never will, especially due to the cheating possibilities. :-)

  • 12
    Only looking for blunders would certainly raise your elo by a non-trivial yet non-egregious amount, but they could probably tell by the fact that you never blunder, whereas people with a similar elo do blunder, they just play better on average, and statistics would tell in the end... just a thought, I don't know for sure actually, I'd rather hear an actual answer from an expert but cheating detection is really sophisticated now
    – Evariste
    Mar 25, 2021 at 23:16
  • 2
    Are you asking if that could be detected in theory (which is likely "yes") or if it can be detected by current cheat detection algorithms? If the latter, this may not be answerable, since platforms like to keep their cheating detection methods secret.
    – Edward
    Mar 26, 2021 at 0:26
  • 4
    I don't see how that's "semi" cheating
    – njzk2
    Mar 27, 2021 at 23:07
  • 3
    Also, are you sure that it “begs” the question? Mar 28, 2021 at 0:02
  • 1
    @leftaroundabout: Begs the question if, as a non-natural speaker, I should use phrases that even natural sometimes botch. Guess I shouldn't :-) Mar 28, 2021 at 10:21

4 Answers 4


Yes. You will still be detected just that the site is going to need more games. You are more likely to stay under the radar, but that doesn't mean you won't be caught. Cheat detection has never been just comparing your moves against the computer.

To understand, online cheat detection is an application of anomaly detection in data science (google "Anomaly Detection" if you need help). Chess.com calculates dozen of metrics such as the probability of beating a higher rated opponent due to luck, and many other metrics that they will never tell you. They are also going to do it for everybody in the database. Once they do it, they will then have a distribution of the metrics for each rating group.

If you are a smart cheater, you will win games slower than a dumb cheater but still more statistically significant than everybody else not cheating. Chess.com will create a mathematical model such as hypothesis testing to derive a p-value probability of you cheating given a null hypothesis that you aren't a cheater. As you win more and more games, the standard errors of your profile will get smaller and smaller, your p-value will also get smaller. It will just be a matter of time until chess.com bans your account.

Even if you don't end up winning, your in-game performance such as your ability to stay consistent with the top-three computer moves or your ability to play time-pressure high complexity positions while not blundering will be statistically different from everybody else in your rating group. Chess.com will be able to detect you even if your moves don't match with the computers.

That's exactly how you see why cheaters like Dewa Kipas were able to cheat for a few weeks until his match with IM Levy Rozman. (https://www.chess.com/news/view/most-watched-chess-stream-in-history-dewa-kipas). Chess.com needed time and data to build up Dewa Kipas's cheating portfolio.

You will be caught given a high enough number of games, but a single game won't be enough.

  • 13
    Surely there must be a point at which sufficiently advanced cheating is indistinguishable from slowly improving your game: if you use randomized moves from a human-like engine to give you a slight gradual ELO increase, for instance. Mar 26, 2021 at 11:25
  • 3
    I'm curious whether "anatomy detection" is a repeated typo. There are lots of results for it on Google, but they all seem to relate to analysis of real anatomical scans (that is, images and sensor data regarding a human or animal body). Perhaps it's a metaphor I am unfamiliar with, or perhaps it was a typo.
    – RoundTower
    Mar 26, 2021 at 11:38
  • 13
    This feels like wishful thinking. Maybe I am misunderstanding? The example given seems to support the idea that cheaters are really hard to catch. As I interpreted the article it was live streaming a real game where the cheater couldn't cheat that outed them. For a casual, low profile cheater, it'd be easy to avoid such a situation. This was a cheater with an ELO over 3000 (with an actual ELO in the 1000s) making thousands of dollars off an in-person game with 12,000 viewers.
    – Rob P.
    Mar 26, 2021 at 20:59
  • 1
    Anomaly detection here is the problem itself, not the solution to it (so I think this answer is borderline "You'll be caught, because there is a system that catches cheaters"). And if the number of games necessary to catch you is large enough, the rating can increase to its new value before you get caught. Mar 27, 2021 at 13:53
  • 2
    @HaukeReddmann: I'm not relieved to learn that a pure cheater can still earn $7000 after and as a result of being unambiguously proven to be a cheater!
    – user21820
    Mar 28, 2021 at 4:53

tl; dr: the stronger you are, the more likely it is that you can cheat in this way and not get caught.

Check out this position:

[FEN "r2q1rk1/5p1p/p2p1bp1/4pn2/2B5/7R/1PPQ1PPP/2B2R1K w - - 0 1"]

This position is from Kasparov-Anand, Las Palmas, 1996. It's White's turn to move, and Kasparov went into a deep think. You might want to think about the position yourself, although if you are not a strong grandmaster, you probably won't see anything special.

At this point Kasparov went into a deep think. Jan Timman started to speculate whether White couldn’t play the very forceful 20.g4. Kasparov’s second Juri Dokhoian immediately confirmed: “That’s what he’s looking at!” Yuri understands Kasparov’s thinking better than anyone else in the world.

Meanwhile White had played 20.Bd5. The game lasted six hours, Anand defended very tenaciously and at around 10 p.m., much to the disappointment of Kasparov, a draw was agreed.

When he left the stage Garry spotted me and walked straight over. “I couldn't win it, could I, Fred?” he asked, with a troubled look on his face. It was a bit shocking: the world champion and best player of all times consulting a chess amateur, asking for an evaluation of the game he has just spent six hours on!

Naturally Garry wasn't asking me, he was asking Fritz. He knew I would have been following the game with the computer. “Yes, you had a win, Garry. With 20.g4!” My answer vexed him deeply. “But I saw that! It didn't work. How does it work? Show me.” He and Anand listened in horror while Juri dictated the critical lines. All of this was captured on video and published in ChessBase Magazine 56 (Feb 1997).

Part five of the series continues:

Most top grandmasters understand all too well how computers can affect the outcome of a game. In contrast to an amateur playing 600 points above his true strength, for whom the computer must dictate practically every move of the game, a strong grandmasters requires only occasional assistance to improve his performance dramatically. There are usually a few critical positions in which a player must decide whether a promising plan can work, or whether it is tactically flawed.


If we return to the example of the Las Palmas game given in part four, Kasparov’s second knew what he was looking at and actually uncovered the solution before the move was played. All that Kasparov needed was one bit of information: “Yes”. He didn’t need to get the message “20.g4! wins” but simply “There is a win” or even, in this specific case “The move we know you are looking at works!” That would have been enough to decide the game.

So yes: if you are a strong grandmaster, then having access to a computer telling you there is something in the position will result in a huge elo boost. You don't even need to know what the move is, you just need to know there is something. Furthermore this kind of cheating will not be easy to detect, since for almost all moves of the game you will do just fine on your own.

On the other hand, if you are an amateur, then you will need help from the computer much more often, and furthermore you'll need specific help. Look again at the position: would knowing that White has a win have helped you at your level? As you mentioned, detecting cheating is by matching one's moves against the computer's. Naturally, once you have to consult the computer more often, cheating becomes easier to detect.

Edit: Quoting from this source on the kind of cheating you propose:

Faced with a complex calculation, a player could sneak their smartphone into the bathroom for one move and cheat for only a single critical position. Former World Champion Viswanathan Anand said that one bit per game, one yes-no answer about whether a sacrifice is sound, could be worth 150 rating points.

"I think this is a reliable estimate," says Regan. "An isolated move is almost uncatchable using my regular methods."

[Regan is an academic who's developing methods to catch cheaters.]

But Regan has ideas on how to catch this kind of cheating. They are untested, but he's confident they'll work. See the source for more.

  • 3
    @SmallChess sure there is more involved, but ultimately detecting cheating is matching moves against computers. I understand that from, e.g., cse.buffalo.edu/~regan/chess/fidelity. If you dispute that statement, please show some evidence.
    – Allure
    Mar 26, 2021 at 2:01
  • 1
    One would just need to look at the ELO improvements to dispute your statement. chess.com/news/view/… has your evidence. The IM in the article was able to spot cheating even before the game by just looking at the ELO graph.
    – SmallChess
    Mar 26, 2021 at 2:04
  • 7
    @SmallChess the IM in the article didn't "spot cheating", he became suspicious of it. To actually prove cheating still involves matching moves against the computer. Please read the link above, or this one for an account written at more accessible level: cse.buffalo.edu/~regan/personal/JuneCLarticleKWR.pdf
    – Allure
    Mar 26, 2021 at 2:18
  • 2
    @HaukeReddmann seems to be more complicate than that: en.chessbase.com/post/the-sunday-treat-garry-at-google The full line: 20.g4! Qc8 (20...d5 21.gxf5 dxc4 22.Qh6 Qd5+ 23.f3 Rfd8) 21.Bd5 Nh4 22.Rg1! g5 (22...Rb8 23.Qh6±; 22...Bg7 23.Rxh4) 23.Rxh4 gxh4 24.g5 Bg7 25.g6 Kh8 (25...Qf5 26.gxf7+ Kh8 27.Bxa8 Rxa8 28.Qd5 wins) 26.gxf7 Qf5 27.Bxa8 (27.Rxg7 Kxg7 28.Qh6+ Kh8 29.Bg5 Rxf7 30.Bxa8=) 27...Rxa8 28.Qd5 Rf8 29.Bh6!
    – njzk2
    Mar 27, 2021 at 23:21
  • 1
    @user541686 I think my initial thought is right then. After 20. g4 Qc8 21. gxf5 (since "I just need that knight to leave") Qxc4 22. Rg1 (if 22. Qh6?? Qxf1#) Qe4+ (defending with tempo) 23. Rg2 Bg7 and White cannot play Qh6. A GM might evaluate all this. Patzers like me might still find the right move, but it would be for the wrong reasons.
    – Allure
    Mar 29, 2021 at 5:47

I'm afraid SmallChess's answer and comments above are more a matter of wishful thinking than actual fact. If you manage to catch a few small-time cheaters every now and then, it becomes easy to convince yourself that you can detect most or all cheaters; by definition, successful cheaters don't get caught, so you can pretend they don't exist. But they do, and perhaps more than you might expect. I know because for a long time i made my living developing and customising bespoke chess engines for this very purpose, and, properly used, over time, engine and (cheating) player become very much indistinguishable. If you think about it, SmallChess is in effect proposing they can produce a robust Turing Test (distinguishing human from machine) that works even in the very restricted context of chess games (as opposed to free-flowing conversation, say). How likely do you think that is?

Believe you me, engine-assisted cheating is almost as widespread as doping in pro cycling.

  • 2
    I didn't say cheating protocol was perfect. I said cheating algorithms required lots of games to make a decision. Simply not matching computer moves will not be sufficient.
    – SmallChess
    Mar 28, 2021 at 2:57
  • It is not just a few cheaters getting caught though, chess.com and lichess ban thousands of cheaters every day... How many do they not find? Well, obviously one can't know, but it's not like just the cheaters are getting smarter, the cheat detectors are too.
    – koedem
    Mar 28, 2021 at 4:02
  • More importantly, is it possible to cheat undetected? Probably. But how many people are going to go through the effort required to become a good enough cheater to not get detected? Probably not all that many since it's a lot of effort. For honestly no real benefit.
    – koedem
    Mar 28, 2021 at 4:23
  • 3
    <shrug> As a scientist, I see the epistemological problem. I wouldn't agree, though, that catching a cheater has the same algorithmic complexity than passing the Turing test. Moreover, as my comment above shows, I relentlessly would have played a computer move that even Kasparov didn't dare - without a computer :-) Any anti-cheater measurement can only be probabilistic, and if only x% go undetected, it suffices to discourage the rest. Mar 28, 2021 at 10:17
  • 1
    Obviously, for detecting cheaters they must have been able to cheat undetected before. Unless your anti-cheating system uses precogs. I would focus instead in the time needed. Do you catch 90% of cheaters after 10 games? 95%? How does it evolve after every added game? How many games do you need to catch 99% of cheaters? And at which point it stops making sense to cheat? A cheat engine that produced the same moves you would perform yourself (like a blackbox asking a chess expert, just that the expert would be yourself) could be undetectable (it would match 100% your expertise) but rather silly.
    – Ángel
    Mar 29, 2021 at 0:03

I hunt cheats on Chess.com and have done it on Lichess also. One thing I can assure is that, even if you are using a machine to assist a human and combining the human's inconsistencies and consistencies, with that of the machine algorithm, the reason the cheater will get caught is that their ability will eventually be mapped and where the artificial intelligence kicks will stick out. Everything you do will leave fingerprints in statistical terms.

I developed my own method for finding bot users specifically, simply because they do present a problem for the typical T-3 analysis methods that are used. They also will frustrate the use of AE detection methods in this way also, but I discovered one thing they can never hide from. It eluded me for many years, but once I discovered it, I laughed. I won't divulge how I do it, no matter what flack I get for not allowing anyone to scrutinize the method. Chess.com isn't going to allow scrutiny of their method either and for good reason. All I can tell you is, I can often see in a single game, even when the T-stats and AE aren't suspicious, and no it isn't move time data. With a sample of 10-20 games though, I can be rather certain if you are a "clever cheater" or bot user.

As for people who want to try to count someone who used an engine once in a great while on a single move in a game as a clever cheater that will never get caught, in my mind such players don't exist. If your ethics are that broken, greed will get the better of you sooner or later.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.