TCEC Keeps AI-Coded Engines Out of Its Top Divisions
Starting with Season 30, TCEC will route engines whose code was generated by AI through a separate qualifying playoff and, eventually, a standalone gauntlet — kept apart from the Leagues, where the championship is actually decided.
TCEC, the Top Chess Engine Championship, said on August 31, 2026 that fully AI-coded engines will not be admitted into its top divisions for the tournament's 30th season. Tournament Director Anton Mihailov announced the policy through Chessdom, the outlet that has organized TCEC for most of its history, laying out a narrow and temporary path for engines whose code was generated by artificial intelligence rather than written by hand.
The mechanism runs through machinery TCEC already had in place. Engines that are new to the event, or returning after time away, are sorted into numbered strength categories, and playoffs among the lower categories have long filled a small number of the limited qualifying spots into the TCEC Swiss, the open stage that feeds the season's top divisions, the Leagues. Under the new rule, the Category 3 + Category 4 Playoff becomes the only door open to a fully AI-coded engine. If one wins it and takes its Swiss slot, the results it posts there go nowhere: Mihailov's statement makes clear an AI-coded engine cannot carry a strong Swiss run into the Leagues, however it performs.
Instead, organizers plan to build AI-coded engines a separate gauntlet of their own — a standalone event, still unscheduled, that keeps their results apart from the Leagues table where the season championship is actually decided. Mihailov framed the split as recognition that "fully AI-coded engines and traditionally programmed engines are fundamentally different approaches," and said the Rules Committee will keep watching the field with the aim of eventually "establishing separate and/or joint competitions" for both kinds of engine.
The policy leaves open a problem it has not solved. On the TalkChess forum, where engine developers discussed the announcement, one commenter, posting as Gabor Szots, asked the question TCEC itself has not answered: how the organization intends to determine whether a submitted engine is human-made, AI-assisted, or fully AI-made, given that entries are typically self-declared. Another poster, Modern Times, called the quarantine "a sensible approach" for now, predicting AI-generated engines "will probably dominate the engine development scene sooner than we think" — a forecast, not a result.
The timing is not incidental. TCEC's Season 29, which ended earlier this year, was itself a story about a monoculture cracking: for years the championship had settled into a two-engine contest between Stockfish's hand-tuned alpha-beta search and Leela Chess Zero's neural-network evaluation, before other approaches broke back into contention. An engine whose source code, not just its evaluation function, was generated by an AI system is a third and distinct category again, and TCEC's response is to wall it off until the Rules Committee decides how, or whether, to score it against the other two on equal terms.
No AI-coded engine has yet played a game under the new rule, and TCEC has not named one that intends to enter. Season 30 is described as the testing ground for the policy itself, not for any specific engine's results — the current state is a rule change, not a competition outcome, and should be read as one.