Story
September 11, 2026
US and Anthropic Say China’s AI Race Is Being Fueled by Model Copying
Washington and Anthropic portray the alleged extraction of frontier-model capabilities as a strategic threat to the US AI lead, while Beijing says the accusations dismiss China’s own technological progress and turn competition into a smear campaign.
US officials say the fight over artificial intelligence has moved beyond competing models and into a battle over the data, reasoning and tools behind them.
The allegations trace back to at least late 2024, when DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI allegedly began industrial-scale efforts to draw capabilities from US systems, according to a joint NSA, CISA and FBI assessment. The agencies said the firms likely operated with Chinese government awareness and targeted variants of Claude, GPT, Gemini and Grok. Their warning was blunt: stolen capabilities form “the core—not merely a supplement”—of China’s AI development strategy.1
The alleged methods were as sprawling as they were technical: bulk-purchased accounts, proxy networks, repeated prompts and jailbreaks designed to expose hidden reasoning. Officials argued that the payoff is enormous—shorter development timelines and lower training costs—but acknowledged that countermeasures could also degrade service for legitimate users. Their proposed answer is tougher monitoring, identity checks, information-sharing and, in some cases, quietly routing suspected attackers to weaker models.1
Anthropic’s subsequent account put scale behind that warning. It said it had identified five campaigns involving nearly 200 million exchanges and attempts to extract Claude’s reasoning traces, coding, data-analysis and tool-use capabilities. Alibaba was attributed 151 million exchanges from May through July 2026, spread over 3,500 accounts; Anthropic called it the largest wholesale distillation effort it had seen. A separate Moonshot AI campaign, it said, routed nearly 300,000 requests through 5,000 accounts in 10 days and included a request connected to surveillance analysis.2
China rejects the premise. Foreign Ministry spokesperson Mao Ning said US agencies should pursue AI cooperation “rather than making groundless accusations,” while the Chinese Embassy described the charges as a prejudicial effort to discredit achievements born of scientific and technological self-reliance.1
The divide is stark: Washington sees systematic theft; Anthropic sees an escalating security problem; Beijing sees a political attack on a fast-rising competitor.