Distillation attacks turn billions in American investment into a subsidy for competitors
In the long contest over who will shape the intelligence of machines, Anthropic has placed before the United States Senate a stark accusation: that Alibaba, through nearly 29 million manufactured exchanges, systematically drained the hard-won capabilities of its Claude AI model — converting years of American research into a rival's advantage at a fraction of the cost. The complaint, addressed to Senators Scott and Warren, frames what might once have been seen as a corporate dispute as a matter of national security, arriving at a moment when the race to dominate artificial intelligence has become inseparable from geopolitical rivalry. It is a reminder that in the age of intelligent systems, the most valuable thing that can be stolen is not data, but the capacity to reason.
- Anthropic alleges Alibaba ran the largest known AI capability theft in history — nearly 29 million interactions through thousands of fake accounts designed to quietly drain Claude's most sophisticated features.
- The technique, known as distillation, allows a competitor to train a cheaper model on the harvested wisdom of a more advanced one, turning billions in American R&D into a shortcut for rivals.
- Anthropic has linked Alibaba, along with BYD and Baidu, to China's military apparatus through Department of Defense assessments, elevating the accusation from intellectual property dispute to national security threat.
- Alibaba has denied military ties, launched legal action to remove itself from the Pentagon's blacklist, and has not publicly addressed the specific distillation allegations.
- US lawmakers now face mounting pressure to impose penalties on distillation attacks and tighten export controls, as Anthropic — itself approaching a landmark IPO — pushes the issue into the political arena.
Anthropic has accused Alibaba of orchestrating the largest coordinated theft of AI capabilities ever documented, alleging in a June 10 letter to US Senators Tim Scott and Elizabeth Warren that operators linked to the Chinese tech giant conducted nearly 29 million interactions with its Claude model through thousands of fraudulent accounts.
The method at the heart of the accusation is known as distillation — a process by which an attacker repeatedly queries a powerful AI system, harvests its responses, and uses them to train a cheaper competing model. Anthropic says the campaign specifically targeted Claude's most advanced capabilities: its handling of complex, lengthy tasks and its distinctive reasoning approach. The company described this as part of a systematic effort by Chinese firms to reverse-engineer American AI at industrial scale, converting what it called hundreds of billions of dollars in US research investment into competitive advantage at a fraction of the cost.
The letter invoked national security, citing Department of Defense assessments that link Alibaba — alongside BYD and Baidu — to China's military apparatus. Anthropic urged Congress to penalize organizations conducting such attacks and to strengthen protections for American technology. Alibaba has denied any military connections and this week filed legal action seeking removal from the Pentagon's security threat blacklist, while declining to address Anthropic's specific allegations.
The complaint arrives as Anthropic prepares for what could be one of the most significant IPOs in the history of the technology industry. Its public escalation signals a broader shift: the battle over artificial intelligence is no longer fought only in research labs and data centers, but in congressional chambers and courtrooms, where the rules governing who owns the mind of a machine are still being written.
Anthropic has accused Alibaba of orchestrating what it describes as the largest coordinated theft of artificial intelligence capabilities ever documented. In a letter dated June 10 and addressed to US Senators Tim Scott and Elizabeth Warren, the San Francisco-based AI company alleged that operators connected to Alibaba conducted nearly 29 million interactions with Claude, Anthropic's flagship AI model, using thousands of fake accounts created specifically to extract its most valuable features.
The attack relied on a technique known as distillation—a method where operators query a sophisticated AI system repeatedly to harvest its responses, then use those answers to train a cheaper, competing model. Anthropic said the campaign targeted Claude's most prized capabilities: its ability to handle lengthy and intricate tasks, and its reasoning approach. The company characterized this as part of a broader, systematic effort by Chinese firms to reverse-engineer American AI technology at industrial scale, converting hundreds of billions of dollars in US research investment into competitive advantage at a fraction of the original development cost.
The letter framed the issue as a matter of national security. Anthropic pointed to Department of Defense assessments linking Alibaba, along with automaker BYD and tech company Baidu, to China's military apparatus. The company urged Congress to impose penalties on organizations conducting such attacks and to strengthen protections preventing American technology from being stolen. "Distillation attacks turn hundreds of billions of dollars in American investment and research and development into a massive subsidy for our geopolitical competitors," Anthropic wrote.
Alibaba has denied any connection to Chinese military interests and has taken legal action against the US government this week, seeking removal from the Pentagon's blacklist of companies deemed security threats. The company has not publicly responded to Anthropic's specific allegations. Other major US AI developers have previously raised similar concerns about Chinese competitors using distillation techniques to build rival systems at significantly lower cost than developing models from scratch.
The timing of Anthropic's complaint is notable. The company, alongside OpenAI, is preparing for what could be a transformative initial public offering that would rank it among the world's most valuable corporations. The letter represents an escalation in public pressure on US lawmakers to act decisively on AI security—a concern that sits at the intersection of intellectual property protection, export controls, and geopolitical competition. As American and Chinese firms race to dominate artificial intelligence, the question of how to defend proprietary technology while maintaining innovation has become urgent for policymakers and the industry alike.
Notable Quotes
Distillation attacks turn hundreds of billions of dollars in American investment and research and development into a massive subsidy for our geopolitical competitors— Anthropic, in letter to Congress
The Hearth Conversation Another angle on the story
Why would Alibaba need to extract Claude's capabilities if they're already building their own AI systems?
Because building a world-class AI model from scratch costs enormous amounts of money and time. If you can query someone else's model millions of times and learn from its answers, you compress years of research into weeks. You get the benefit of all that American investment without paying for it.
But how do you extract something from an AI model just by asking it questions?
You ask it thousands of variations of the same problem. You watch how it reasons, what it prioritizes, how it handles edge cases. Then you feed all those question-answer pairs into your own model as training data. Eventually your model learns to mimic the original one's behavior.
So it's not stealing code or data—it's learning by observation?
Exactly. Which makes it harder to detect and harder to prosecute. You're not breaking into servers. You're just using the service the way millions of legitimate users do, except at massive scale with fake accounts.
Why does Anthropic care so much right now? They're about to go public.
Because they need to show investors and regulators that their technology is defensible. If Chinese competitors can freely extract your capabilities, your competitive moat disappears. The letter is partly a warning to Congress, but it's also a signal to the market: we take security seriously.
What happens if Congress doesn't act?
The cost of building AI keeps dropping. American companies lose their advantage. And the geopolitical balance shifts—whoever controls the best AI systems has enormous leverage over everything from military capability to economic power.