Sophisticated Cheating in Chess

The chess world was set ablaze this past week with a news release from Chess.com:

Tigran L. Petrosian (a 2500+ rated GM currently ranked #260 in the world, not to be confused with the former world champion sharing the same name) has been found violating fairplay rules by Chess.com’s anti-cheating crew both in the semifinal and final of the PRO Chess League last week-end. The hosting platform and the league have banned the Armenian grandmaster for life. The league has also disqualified his team, the Armenia Eagles, and stripped them off their ($20,000) prize money.

Hikaru Nakamura is currently ranked #4 in the world in rapid chess and has the largest Chess stream on Twitch. He recently uploaded a YouTube video with his thoughts on the cheating scandal and potential ways the chess world can move forward in the face of cheating scandals.

The Security Mindset

The recent drama with Tigran Petrosian banned from Chess.com after obviously cheating at a 2900 performance level with associated eye movements from his web camera is very interesting.

It is obvious that players playing every single move at engine strength or playing at a 2900 performance rating are easy to detect. But what if the cheaters are more sophisticated?

Security expert Bruce Schneier describes the Security Mindset as framing a problem through the eyes of an attacker.

Good engineering involves thinking about how things can be made to work; the security mindset involves thinking about how things can be made to fail.

It involves thinking like an attacker, an adversary or a criminal. You don’t have to exploit the vulnerabilities you find, but if you don’t see the world that way, you’ll never notice most security problems.

Individuals with a security mindset can’t walk into a store without thinking how they might shoplift. They can’t vote without thinking about how they might vote twice.

Security Theater in Chess

Schneier describes Security Theater as security measures that make people feel more secure without doing anything to actually improve their security.

I venture to suggest that most of the methods currently being suggested to combat cheating in online chess tournaments with cash prizes is merely security theater that will only detect the lowest hanging fruit. Sophisticated cheaters will go undetected and enjoy the cover this security theater provides them.

Some of the methods currently beginning to be employed and supported by many of the top chess players include:

1. Webcams — possibly including multiple webcams from different angles around the player
2. Live proctors
3. Audio recording
4. Screen sharing

Using the security mindset I will describe some techniques a sophisticated cheater could use to defeat all of these methods.

Statistical analysis is the most reliable method of detecting cheating and in some cases can identify computer assistance with a stronger confidence than DNA testing that we rely on to send countless murderers to prison.

But what about when a motivated, sophisticated individual who is familiar with statistical analysis wants to cheat? What if esteemed anti-cheat professor and international master Dr. Ken Regan wanted to defeat his own system? Do you think he could defeat his own detection? I venture to say the answer to that question is Yes.

Reputational Protection Fallacy

Hikaru Nakamura mentioned that the risk to reward for top players who have invested an entire career in chess is just not there.

We know from statistics released by Chess.com in addition to the recent scandal this just isn’t true. While this is a strong argument for the world top 10 — it much less reliable for the overall chess community.

Chess.com disclosed they have closed 300+ titled players accounts for fair play violations including at least 3 2600+ GMs. They have confidence in their statistical analysis and are willing to defend their bans in court. Many of these titled players have signed written confessions and chess.com agreed to conceal their grievances in their amnesty program.

This is compelling evidence in favor of statistical analysis but I argue that these players did not use sophisticated methods and became low hanging fruit for the anti-cheat algorithms.

Defeating Webcams

The easiest method to defeat webcams is to use an accomplice. This accomplice doesn’t need to be a chess player. More sophisticated automated solutions leveraging the same methods to send and receive signals could also be leveraged by a talented programmer but the simplest way is to have an accomplice.

The webcams are also often feeding to a Zoom call. A program could intercept the video feed on the way to the Zoom call and hide the portion of a screen showing an engine if the player wants to cheat without an accomplice.

The accomplice needs to review the moves and transmit signals to the player. The accomplice could also be trained in statistical analysis and not even tempt the player with moves that would raise red flags.

First, the accomplice first needs access to the moves of the player. This could be accomplished in many ways directly or even just via live tournament feed of the game.

Some examples include:

1. A remote access program mirroring the players screen
2. Plugging a second monitor into the computer from an adjacent room

Next the accomplice needs to be able to transmit moves to the player. Some examples of potential methods include:

1. Haptic sensor or air pad placed on the players leg with an accompanying RF receiving device in underwear
2. Microscopic earphone (check your local spy shop) placed touching players eardrum

A hobbyist with a spare weekend could engineer these systems or custom software to hide a cheat engine from a Zoom call is one $300 freelancer.com project away for a motivated cheater.

Defeating Statistical Analysis

Defeating statistical analysis is a much more difficult problem than defeating a webcam. If you play engine perfect moves and at a 2900 performance rating like Tigran Petrosian you will quickly be discovered via statistical analysis just like the naïve 300+ titled players banned from chess.com with irrefutable statistical evidence.

Applying the security mindset statistical analysis can also be defeated. No doubt this will be a never ending battle and tournament organizers and chess platforms will develop more sophisticated anti-cheat methods with time — most all relying on statistical analysis.

Just because the problem is difficult doesn’t mean a motivated cheater can’t accomplish his goal.

A motivated cheater or accomplice can become an expert at using an open source tool like Irwin — the tool used by Lichess.org to remove cheaters and the evolution of CheaterNet. While chess.com may have more sophisticated algorithms for detection — their stance on irrefutable proof will be a difficult burden to achieve for a cheater well versed in defeating Irwin.

Using Irwin the cheater will select a targeted performance rating. Even a performance rating as low as 2000 is enough to win via just making blunders that don’t cost you games such as when you are already in a winning position.

Select large samples of games from honest players at your desired performance rating. Select large samples of games from closed accounts on Chess.com. Feed the games through Irwin and learn the patterns it uses to differentiate honest players from cheaters.

Strive to emulate those patterns in your own games. Make many online “throwaway” accounts where you practice your cheating methods and then feed the games through Irwin — identifying and eliminating patterns it uses to detect cheaters. Eventually with enough practice a sophisticated cheater can use his engine in a way that always has plausible doubt via statistics.

Technique Brainstorm to Hide Cheating From Statistical Analysis and Human Suspicion

An experienced cheater practiced with Irwin feedback would likely come up with many more advanced techniques to hide their engine use. Some potential techniques might include:

1. Use a weak engine — 2000 is strong enough to beat most GMs when used in critical moments, especially in rapid chess as it will see all forcing lines
2. Use a custom engine that intentionally includes patterns from your statistical practice via Irwin (another $500 freelancer.com project away for a motivated non-programmer)
3. Mimic techniques used by developers of bots like HikaruBot meant to simulate a real player
4. Make intentional blunders when its unlikely your human opponent will find the punishment (computer looking move is only punishment) to reduce your performance rating
5. Only use the engine to make forcing moves like tactical sequences beginning with a capture, make long pauses and calculate random variations in your head while intensely staring at the screen in the critical position if on webcam or playing over the board
6. Intentionally overlook several winning tactics without giving opponent any chances until eventually playing an obvious tactic a human could reasonably often find
7. Create an opening book consisting of large selection of random Expert-IMs from the database to create a “believable” opening repertoire. This way your opponents aren’t suspicious when you play like trash in opening than miraculously outplay them in middlegame
8. When in a winning position “slow down” the win — let your +5 advantage drop to +2 and only play obvious continuations
9. Lose games in proportion to your targeted performance rating via your Irwin analysis
10. Play using your modified engine style in low stakes tournaments and intentionally lose while noting the 1–2 critical positions you would win with in a high stakes tournament
11. Vary time spent per move in accordance with Irwin study
12. Only use the engine for 1 critical moment per game maximum, possibly via an accomplice trained to select this or via practice or via custom software trained to only feed you 1 critical move per game
13. Purchase an anonymous Lichess or Chess.com account at your desired performance rating and attach your name to it. If questioned about your performance at an over the board tournament show them the account with years of online history. This account’s history could also be used to craft your opening book that will help avoid human suspicion from similar rated players.

Conclusion

Sophisticated, motivated cheaters definitely exist among the top 500 players in the world. 300+ titled players that made themselves low hanging fruit playing at a 2900 level have shown the risk is worth the reward to many.

It is foolish to think that some with the intellect to achieve an IM or GM title would not have the intellect to employ techniques outlined above.

Cycling, track and field, football, baseball, poker, bridge and many other competitive activities have all had their share of cheating scandals. When money and ego are involved cheating is always present.

It is much easier and more effective to cheat at chess than many of these other activities. It is definitely happening and will not be an easy problem to combat — especially online.