AI Solves a 58-Year-Old Chess Problem: 63 Maximum Checks and a Bound Still Open

On the night of August 3, 2026, at the National Stadium in Tokyo, I sat replaying slow-motion footage twenty times. Karsten Warholm had just run the 400-meter hurdles in 46.70 seconds, breaking the world record. I counted every stride between the hurdles, not to confirm the time, but to understand how a human body could arrange hundreds of landings into a near-perfect straight line. In nearly forty years in this trade, I have always gone looking for the structure behind the numbers people celebrate. Then one afternoon in Da Nang, I came across an announcement that made my hand stop mid-page. An artificial intelligence system called Double Agent announced it had proven the maximum number of legal checking moves in a single chess position is 63. The previous human record was 60. At the same time, the system proved an upper bound of 364 consecutive checking moves, while the position built by humans in 2026 reached only 54. What kept me sitting there was not the machine beating humans. That story is old. What kept me sitting there was the way Double Agent reached its conclusion: it did not search, it proved. And between the two numbers, 63 and 364, there is a gap that the headline "AI solves a chess problem" has blurred away. To understand why this story deserves a sports writer's time, we need to return to a very narrow corner of chess that few Vietnamese audiences know: combinatorial chess, also called chess problems. This is the branch where people do not compete in tournaments, there is no Elo, no clock. People compose positions with a specific technical requirement, and the community verifies whether the position satisfies it. The problem Double Agent solved belongs to the harshest type in this branch: find a position in which the side to move has only checking moves. In other words, every legal move available must give check, and at least one legal move must exist. That seemingly simple requirement becomes an enormous combinatorial problem, because every piece, every empty square, every diagonal, every file can neutralize a whole set of other moves. The story has run for 58 years. If the starting point is placed in 2026, the problem has now passed through nearly six decades and several generations of composers. People like Norbert Ringeltaube once held the record of 60 checking moves. People like Alexey Khanyan in 2026 built a position with 54 consecutive checking moves. They are not famous like grandmasters, they never appear on any tournament scoreboard. But they are the ones who pushed the limits of the game further. That is why I could not ignore this story. Pulling back the curtain on speed, I found a whole generation running. Here, that generation does not run on a track but inside black and white squares, quietly for more than half a century, and finally recognized by a machine. We need to be clear from the start about the nature of this announcement, because this is the point most news items skipped. This is not a match. No player sits at a board. There is no tournament, no prize money, no grandstand. Double Agent did not defeat anyone. It proved a combinatorial theorem within the framework of chess rules. Understanding this, the reader will see that the headline "AI solves a 58-year-old chess problem" carries two very different layers of meaning. The first, narrow and precise: a specific extremal problem in a narrow branch has been solved. The second, broad and easily misread: artificial intelligence has conquered chess. The distance between these two layers is where the real story lies. So what did Double Agent do? According to the research team's description, the system did not plunge into searching the entire space of positions. The team asserts that approach is computationally infeasible. Instead, they took another path: first prove structural and geometric properties of checking positions, then formalize the entire argument in Lean 4, a mathematical proof assistant, then reduce the remainder to a SAT problem, and finally let the machine check every step. This is the core point I want to stress: the real achievement lies in the method, not in the number. Turning a chess problem into a chain of formal, machine-checkable arguments is a leap in kind. It transforms the claim "I found a position with 63 checks" into a much stronger claim: "no position can exceed 63 checks." Those two claims differ by an ocean. One can build a position with 63 checks by trial and error, but to prove that no one can do better, one must block every possibility. That is why this problem has existed for 58 years. Humans can always find something better, but proving the best is another matter entirely. According to the published information, the 63-check position is not merely a theoretical construct. The research team asserts it can arise from a legal game. This detail matters more than it appears. In the world of chess composition, the boundary between a "real" position and a "fairy" construct is a matter of honor. A position is only beautiful when it could occur in an actual game, where both sides follow the rules from the first move. To understand why the legality requirement is so hard, we need to look at how chess rules define a move. A move is legal when it does not leave one's own king in check. A pinned piece cannot move if that move exposes the king to attack. A square controlled by another piece cannot be a destination. And a checking move is one that directly attacks the enemy king. When all these constraints are combined, the problem becomes a labyrinth of logic. To maximize the number of checking moves, the composer must arrange pieces so that every legal move of the side to move is a check. That means blocking every non-checking move, without blocking so much that the side runs out of moves. If the side to move has no legal move left, the position falls into stalemate, and the problem collapses. From my analysis, the 63-position almost certainly must achieve extreme piece density and use pins and line-blocking to suppress every non-checking move. The details mentioned, such as "tight arrangement," pins, line-blocking, and square control, are all consistent with this hypothesis. When the board is crowded with pieces, each piece is both a threat and an obstacle, and the composer must balance those two roles on every square. This is where I remember the evenings I built the video series "Pulling Back the Curtain on Speed" in 2026, when Nguyen Thi Oanh won the 1,500-meter gold at the SEA Games 29 with a time of 4 minutes 15.07 seconds. I analyzed each of her laps using replay software. What I learned then still holds here: to understand a performance, you must break it into its smallest units, then see how those units fit together. With a chess problem, the smallest unit is one legal move. When we count 63 checking moves, we understand nothing yet. Only when we see why there cannot be a 64th do we truly touch the essence of the problem. And that is exactly what Double Agent is said to have done. Now comes the most important part, the part short news items blurred. The research team published two results, but those two results have completely different statuses. The first result, 63 checking moves, is a closed result. It has both a lower bound and an upper bound. The lower bound is constructing a position reaching 63 moves. The upper bound is proving no position can exceed 63. When the two bounds meet, the problem is considered solved. This is a complete extremal theorem. The second result, 364 consecutive checking moves, is only half a result. The research team states clearly that 364 is an upper bound. That means no position can have a consecutive checking sequence longer than 364. But that upper bound does not come with a position reaching 364. The true number remains open. This difference is not a trivial matter of wording. It determines how we assess the entire announcement. For the first problem, we can say: solved. For the second, we can only say: the space has been narrowed. The gap between the 54 moves Alexey Khanyan built in 2026 and the 364 the machine proved as an upper bound is still vast. A skimming reader will see the headline "AI solves a chess problem" and think everything is settled. In reality, half the story is still open. This is the kind of confusion I have witnessed many times in my career, when an impressive number is detached from its context and begins to live a life of its own, larger than the original truth. One more technical point deserves mention, which the research team offered as an observation, not a proof. They suggest that adding pieces or increasing attacking power may not improve the result. This is an opinion, and it is presented as one. In a field where every claim must come with a formal proof, distinguishing a theorem from a conjecture is a survival skill. At this point, I want to turn to another angle, one I consider more important than the number. In recent years, I have written repeatedly about how heat maps and statistical indices are gradually becoming a new kind of fortune-telling in sports. People look at a colorful chart and think they understand a player. But the chart often hides the person's real role within a tactical system. The Double Agent story carries the opposite lesson, and that is what makes it worth pondering. Here, the machine does not offer a vague prediction. It offers a claim checkable step by step. The difference is this: a heat map says "it seems," while a formal proof says "certainly." In an age flooded with vague data, the ability to distinguish these two kinds of claims is a precious skill. But I also do not want to deify the machine. There is a question this announcement leaves open, and it deserves to be stated plainly. The proof is called "machine-checked," but the checking machine belongs to the research team itself. No third party is mentioned in an independent audit role. What does this mean? It means "machine-checked" only reduces the risk of error if the formalization process itself is correct. If there is a gap in how the problem is encoded into SAT, or an unchecked assumption in Lean 4, the upper bound could collapse. This is the classic failure mode of formalized extremal claims: a correct proof, but a proof of a misdefined problem. I am not saying this happened. I am saying it has not been ruled out by an independent source. And in my trade, the difference between "not yet refuted" and "confirmed" is the difference between a rumor and a fact. There is one more detail I want readers to notice. The original article I relied on does not specify its outlet. Most facts are not attributed to any specific source. Only the headline numbers are attributed to Double Agent. This forces me to write this piece with a certain caution, and I advise readers to do the same. This caution does not diminish the story's value. It only places the story correctly. This is a technical milestone in a small community, not a revolution shaking the world of chess. The accurate phrasing would be: an AI system proved a narrow extremal theorem in combinatorial chess, and proved an upper bound for another problem still open. That is the sentence I want people to quote, instead of "AI solves a chess problem." So what is truly valuable here? I believe the value lies in Double Agent standardizing a workflow. That workflow has four steps: prove structural properties, formalize in Lean 4, reduce to SAT, and machine-check. If this workflow is accepted by the chess composition community, it could raise the verification standard for the entire branch. Future extremal claims will no longer rest on personal reputation, but on proofs anyone can re-check. I once witnessed something similar in a completely different field. When the podcast "Empty Stands" was born during the pandemic in 2026, I had to learn to listen to a match without a crowd. The first episode was a call with swimmer Nguyen Huy Hoang, from Quang Binh, who had to train in the Gianh River when pools closed. No cheering, no electronic scoreboard. Only the structure of things remained. Empty stands, but the podcast taught me to listen to the match from within. This chess problem is the same. When we strip away the glare of the sensational headline, when we strip away the noise of the word "AI," what remains is a pure structure: a problem, a proof, and a limit not yet reached. There is a human dimension I do not want to skip. Behind the two numbers 60 and 54 are two real people. Norbert Ringeltaube and Alexey Khanyan are not stars. They are quiet composers who spent years of their lives pushing a technical limit a few steps further. When a machine arrives and surpasses them, the story is easily told as: humans lost to the machine. But that is the wrong telling. The right telling would be: humans posed the problem, humans built the first records, humans designed the machine, and humans wrote the proof the machine re-checks. In that entire chain, there is no step where humans are absent. The machine did not come from nowhere. It came from the generations who sat counting checking moves for 58 years. This is what I always try to remind myself in every piece. People talk about speed, but I find time. People talk about a record broken, but I find the people who built the old record. In this story, the time is 58 years, and the people are the composers who kept the flame for a branch of chess most audiences do not know exists. From the sandpit track to the virtual arena, the homeland of sport has no borders. I have traveled from the athletics tracks in Kuala Lumpur, through the volunteer area at Luzhniki stadium where I met Ahmed, to the National Stadium in Japan on the night Warholm broke the record. Each time, I saw the same thing: humans trying to push some limit a little further. This chess problem is another expression of the same instinct. What is fascinating is that the instinct does not distinguish forms. A person running the 400-meter hurdles and a person composing a 63-check position are doing the same thing: finding a way to arrange elements so that efficiency reaches its maximum, while obeying unbreakable constraints. Warholm is bound by gravity and footwork technique. The chess composer is bound by chess rules and the legality requirement of the position. For this reason, I do not treat this problem as alien to sport. I treat it as a branch of the same current. In the silence of the stands, the future puts on running shoes. But sometimes the future does not wear shoes. Sometimes it sits down, picks up a pen, and composes a position no one ever imagined. Now let us talk about what this announcement can change, and what it cannot. It can change how the chess composition community verifies extremal claims. If the Lean 4 plus SAT workflow becomes standard, claims like "I found a new record" will need machine-checkable proofs attached. This is a cultural change, not merely a technical one. It shifts the focus from reputation to evidence. It can also create a little attention for a branch of chess very few know. A 63-check position has the potential to become an interactive puzzle on online chess platforms. But according to what has been published, no platform has announced a partnership. This is a possibility, not an event. Commercially, the impact is near zero. There is no prize, no sponsor, no event to exploit. The entire commercial footprint of this story lies in content and novelty. And in an age where everything is measured in views, that is a notable point. For public image, the impact is double-edged. On one hand, it reinforces the narrative "AI is advancing in chess." On the other, it is easily over-read as "chess has been solved." The second direction is the dangerous one, because it creates a false expectation about the real capabilities of artificial intelligence in this field. At this point, I want to return to the central question: is this story sustainable? My answer is: sustainable if read narrowly, and fleeting if read broadly. Read narrowly, we have a closed extremal theorem about 63 checking moves, plus an upper bound about 364 consecutive checking moves. Read broadly, we have an attractive but vague headline about AI conquering chess. The distance between those two readings is where I want to stop. Because in nearly forty years writing about sports, I have seen too many times a small truth inflated into a large legend. And each time, the real people, the real work, the real human beings are hidden behind the halo of the number. This time, I do not want Norbert Ringeltaube and Alexey Khanyan hidden. I do not want the research team behind Double Agent hidden. And I do not want the truth that half the problem remains open hidden behind the neat headline "AI solves a chess problem." In a community I have followed for many years, the value of a position does not lie in whether it shocks. It lies in whether it is correct. A beautiful position is one where every move has a reason, every piece has a role, and no detail is superfluous. Double Agent, if what has been published is accurate, followed exactly that principle in how it worked. So instead of asking "can AI solve chess," I want us to ask another question. When a machine can prove limits humans cannot prove, what will the human role in this game be? The answer, I believe, is already embedded in this very story: humans pose the question, humans build the first records, humans design the machine, and humans decide the meaning of the result. The machine does not ask. The machine only answers. And the best answer it gave this time can be summed up as follows: the maximum number of legal checking moves in a single chess position is 63, no more. That is a complete answer to a question that has existed for 58 years. The question about the 364 consecutive checking moves is still waiting. It waits for a position no one has built, and perhaps, for a person no one knows yet. As I left that page on that afternoon in Da Nang, I thought of the Vietnamese children learning chess in small clubs. Perhaps one of them will grow up and build a position reaching 364 consecutive checking moves. Perhaps not. But what is certain is this: every time a limit is proven, another door opens. And in the world of chess problems, the door always opens forward. I learned this from the quietest people in the trade. Chess composers have no audience. They work in silence, sometimes for decades, to push a number one unit further. When the machine arrives, it does not deny their work. It confirms that the path they walked was right, that the limit they sought is real, and that there is a final stopping point somewhere ahead. That is why I wrote this piece. That is why I believe sport, whether on the track or on the board, whether with muscles or with algorithms, is always the story of humans trying to understand themselves. And in that story, every proven limit is one more time humans see more clearly the nature of the game they are playing.