Does China’s data advantage still mean what it meant five years ago — a billion-person surveillance state feeding proprietary troves into domestic models — or has it quietly become something else: the ability to query US frontier models millions of times and distill their outputs into a competitor for free? The evidence assembled by outside researchers this year points hard toward the second answer. Council on Foreign Relations fellow Chris McGuire describes how Moonshot AI’s Kimi K3, released July 16, was trained partly on “data illicitly obtained from leading U.S. AI labs” via distillation attacks, with the White House alleging banned chips located in Thailand were also used — a mechanism that has little to do with China’s own data lakes and everything to do with harvesting someone else’s.
That distinction matters enormously for anyone pricing data rights. If a lab can back out a meaningful share of another company’s training process by hitting its API hundreds of thousands of times, then the market value of that output — the thing licensing deals and annotation contracts are supposed to protect — is being arbitraged away outside any contract at all.
An advisory body warning about China’s data stockpile is fighting the last war if the real leak is happening through API calls, not datasets.
Treasury Secretary Scott Bessent has called this practice “IP theft,” per CFR’s account, which is a stronger label than “competition” but doesn’t come with an enforcement mechanism. Meanwhile Foreign Affairs’ Colin Kahl argues the more durable US edge isn’t data or even models but compute — and Washington just loosened chip export controls to sell Nvidia’s H200 into China, a decision that cuts against any narrative of clean bifurcation. For data brokers and annotation vendors watching Washington fret over China’s data pile, the more useful question is whether policymakers will start pricing distillation risk into export and API-access rules, because that’s the channel actually moving capability across the border right now.
US advisory body says China's data dominance gives it AI advantage
— Reuters
