Story
September 11, 2026
Anthropic’s China Distillation Claims Sharpen an AI Race Already Turning Hostile
Washington and Anthropic portray industrial-scale model distillation as a threat to America’s AI lead, while Beijing rejects the accusations as politically driven and credits China’s progress to its own scientific and technological capacity.
Anthropic had already warned in February that frontier-model distillation was becoming a serious problem. By Thursday, it said the activity had escalated: China-based labs had developed “increasingly sophisticated methods” to evade defenses and harvest capabilities from U.S. systems, including Claude’s coding, analysis, tool-use and reasoning functions.1
The company’s account put scale behind that warning. Anthropic said it had linked nearly 200 million exchanges to five campaigns. Its largest alleged operation, attributed to Alibaba, involved 151 million exchanges between May and July 2026 across 3,500 accounts, all using a fixed prompt designed to draw out chain-of-thought material for Qwen-model training. It also alleged that a Moonshot AI campaign routed requests from the Chinese military through thousands of accounts.1
The U.S. government had raised the stakes days earlier. In a joint Tuesday release, the NSA, CISA and FBI named DeepSeek, Moonshot, Alibaba, MiniMax, StepFun and Z.AI, alleging they had targeted Claude, GPT, Gemini and Grok since at least late 2024. The agencies said such extraction could shorten development timelines and cut the cost of training rival frontier models, calling for industry-wide defenses against “aggressive, malicious, and targeted distillation activities.”2
Their proposed response—tighter identity checks, monitoring of anomalous accounts and, in some cases, quietly degrading responses to suspected attackers—also exposes a fault line: protections aimed at hostile campaigns could frustrate legitimate users or reduce service quality.2
Beijing’s response was blunt. Foreign Ministry spokesperson Mao Ning said U.S. agencies should pursue AI cooperation “rather than making groundless accusations,” arguing that China’s fast-growing AI sector reflected “high-level scientific and technological self-reliance.”3 In Washington, however, the political mood is moving the other way: David Sacks amplified the view that there is “simply no basis” to believe China would pause if the U.S. did.
4