Anthropic Urges Crackdown on China's AI Distillation — 28.8 Million Exchanges with Claude
By John Nada·Jun 25, 2026·4 min read
Anthropic alleges Alibaba's mass AI distillation attack on Claude, urging Congress for stricter measures. Scale: 28.8 million exchanges via 25,000 accounts.
When Anthropic cried foul, the game changed. Allegations flew that Alibaba-affiliated operators were not just playing, but distilling—extracting AI model capabilities wholesale. According to Decrypt, this wasn't a small skirmish; it was the largest known AI model distillation campaign to date.
Anthropic's clamor for Congress to act resonates with urgency. Their plea: strengthen export controls, bolster intelligence sharing, and penalize firms engaged in such large-scale AI model extraction. The call to action arrived as lawmakers mull over targeting unauthorized access to U.S. frontier AI models. In a letter to Senate Banking, Housing, and Urban Affairs Committee Chairman Tim Scott and Ranking Member Elizabeth Warren, the drama unfolds.
The letter, dated June 10, details the alarming scale of the campaign against Claude, Anthropic's chatbot. From April 22 to June 5, more than 28.8 million exchanges were logged. Nearly 25,000 "fraudulent accounts"—entities detached from real users—were said to be involved, as Anthropic laid out the staggering details. These operations targeted critical capabilities like agentic reasoning, software engineering, and long-horizon planning. By replicating these advanced model behaviors, competitors could bypass the hefty training costs typical of frontier AI systems.
Anthropic's portrayal of the campaign as brazen is not without reason. Alibaba, after all, is a New York Stock Exchange-listed entity with U.S. business operations, making it accountable to American investors and regulators. But the stakes extend beyond intellectual property. The national security angle looms large, as such distillation could catapult China's military and cyber capabilities, while narrowing the technological lead of the U.S.
The situation underscores a pivotal moment as Washington ramps up efforts to shield American AI superiority. Just weeks ago, President Donald Trump signed an executive order to enhance AI-powered cybersecurity initiatives, though it was initially stalled over fears of compromising America's edge against China. The economic inversion Anthropic describes is stark—American investments transforming into subsidies for Chinese competitors.
So what's next? Anthropic advocates for deeper intelligence sharing between frontier AI developers and the U.S. government, clarifying antitrust policies to enable information exchanges about distillation attacks. Export controls on AI chips need tightening, and loopholes that grant Chinese firms access to overseas data centers must close. Penalties for large-scale model extraction, they argue, are not just warranted, but essential.
Though Anthropic's spokesperson declined to comment specifically on the letter, their stance is clear. Fighting illicit distillation is a joint venture between government and industry, a narrative they intend to champion with Congress and the administration.
The letter isn't an isolated incident. Anthropic had sounded the alarm back in February, accusing Chinese AI developers like DeepSeek, Moonshot AI, and MiniMax of similar tactics. That episode saw more than 16 million exchanges with Claude, facilitated by around 24,000 fraudulent accounts. Critics, however, argue that AI firms often use similar methods themselves while developing systems.
Yet, Anthropic distinguishes legitimate distillation for producing smaller models from unauthorized extraction of frontier capabilities, which breaches service terms. This broader debate over distillation isn't new. In April, Elon Musk's testimony in federal court revealed xAI's partial use of OpenAI models in training Grok, highlighting industry practices and the contentious line between legitimate model training and unauthorized extraction.
As Anthropic's letter makes clear, the implications of these distillation practices reach far beyond commercial competition. By effectively siphoning off the fruits of American research and development, these practices threaten to undermine the foundations of U.S. technological leadership. Such activities risk turning billions of dollars of U.S. investment into what Anthropic describes as 'a subsidy for our competitors.'
Yet, the response to these concerns must be as nuanced as the problem itself. Anthropic calls for a multi-faceted approach: expanding intelligence sharing, refining antitrust rules to facilitate cooperation among U.S. AI firms, and closing loopholes that allow unauthorized access to American AI advancements. It's a delicate balance of maintaining competitive edges while ensuring national security interests are not compromised.
This ongoing battle over AI distillation is not just a technological challenge but also a diplomatic and economic one. As the U.S. government and AI industry leaders like Anthropic work together to address these issues, the world will be watching closely to see how this evolving narrative will shape the future of AI development on a global scale.
With the stakes set high, the need for swift and decisive action remains critical, ensuring that the technological strides made by American companies do not inadvertently strengthen foreign competitors. As Anthropic continues to advocate for significant policy changes, the dialogue surrounding AI distillation is likely to intensify, highlighting the urgency of collaborative efforts to protect U.S. innovations from unauthorized exploitation.