Ke Jie reportedly found a weakness in some Go AIs, but its current validity is unproven

The claim comes from a widely circulated video and social-media discussion. Its proposed tactic is to make a sequence of seemingly irrational moves early in the game, causing a Go AI to misjudge the position and creating opportunities to exploit its subsequent mistakes.

The available material does not identify the tested AI, software version, board settings, or a reproducible test procedure. It also does not independently establish that the tactic works against current leading Go engines. The suggestion that the same method would have defeated AlphaGo is an unverifiable counterfactual, not an established result.

The report is therefore best treated as a discussion of adversarial robustness and gaps in an AI system’s training distribution, rather than confirmation of a universal way to defeat Go AI.