Computer scientist Daniel Keren from University of Haifa measured cheating effectiveness in chess by testing intervention algorithms between chess engines. Limited (1-4 moves) engine assistance can increase win-rate to 67-91%, compared to 51% with honest play.

Keren allowed engines to ask a for a help, within limited budget, from better Stockfish engine during different times of the game. Random interventions proved the least effective at 51.2% winrate, providing minimal improvement over baseline play.

Oracle method. This theoretical approach determined the best possible interventions after the game ended, achieving perfect scores but remaining impossible to implement in real play.

Fixed thresholds method. This approach achieved the highest performance by intervening when the difference between strong and weak moves assessment exceeds specific values. One intervention yielded 65% average win-rate, two - 76%, and three interventions achieved 84% win-rate.

Maximal delta predictions (AI approach). The algorithm calculated expected position improvements to determine intervention timing. Bayesian optimization across thousands of games identified optimal intervention points. It took more than 1 month to calculate 3 interventions using powerful chess-server.

Chess game is very fragile. A match may contain 30-40 normal moves, but its outcome often hinges on one critical position. Computer assistance at that single moment negates all previous honest play by the opponent. Future anti-cheating systems should analize move importance rather than overall game accuracy.

Alexander Grischuk recently said that cheating is like murder in criminal law. In March 2025 GM Kirill Shevchenko received a three-year ban after admitting to hiding a phone in a bathroom during the Spanish Team Championship 2024. Chess.com closes approximately 125,000 accounts monthly for fair play violations.

Credit: Lennart Ootes/Grand Chess Tour