On June 10, 2026, Anthropic sent Senators Tim Scott and Elizabeth Warren a letter accusing operators linked to Alibaba and its Qwen AI lab of running the largest illicit extraction campaign against Claude to date. Beyond the accusation, the case shows how distillation has become an industrial-scale threat that Anthropic now wants Congress to address.
🔑 Key Takeaways
- Anthropic alleges more than 28.8 million exchanges with Claude, generated through nearly 25,000 fraudulent accounts between April 22 and June 5, 2026.
- The company calls it the largest distillation attack it has identified, targeting agentic reasoning and software engineering.
- Alibaba rejects any military affiliation; the allegations remain unverified claims at this stage.
- Anthropic asks Congress to update antitrust rules, tighten chip export controls and impose penalties on Chinese labs.
A Campaign of Unprecedented Scale
According to the letter, the campaign ran from April 22 to June 5, 2026. The operators allegedly generated more than 28.8 million exchanges with Claude using nearly 25,000 fraudulent accounts. Anthropic describes the operation as “the most extensive campaign ever measured” and “the largest known distillation attack targeting Anthropic to date.” The company says the activity violated Claude’s terms of use and access restrictions.
Distillation is a training method that builds a smaller or less capable model from the outputs of a more powerful one. Anthropic believes such attacks are now carried out “at industrial scale,” allowing Chinese companies to harvest and repackage American capabilities as their own. In the company’s account, Alibaba’s goal was to extract those capabilities “without bearing the training and research and development costs” needed to build its own frontier model.
Anthropic says the campaign targeted specific strengths: agentic reasoning, software engineering, long-running tasks, handling longer and more complex problems, and decision-making. These are precisely the areas where Western labs are now trying to differentiate themselves.
A Pattern of Distillation Campaigns
The accusation builds on earlier disclosures. In February 2026, Anthropic identified three distillation campaigns at DeepSeek, Moonshot and MiniMax, which it said were growing in intensity and sophistication. It reported more than 16 million exchanges through about 24,000 fraudulent accounts and called for cooperation among AI companies, cloud providers and policymakers.
| Campaign | Exchanges | Fraudulent accounts | Named actors |
|---|---|---|---|
| February 2026 | More than 16 million | About 24,000 | DeepSeek, Moonshot, MiniMax |
| April 22 to June 5, 2026 | More than 28.8 million | Nearly 25,000 | Operators linked to Alibaba and Qwen |
The second campaign would therefore involve nearly twice as many exchanges as the first, with a comparable number of accounts. OpenAI and Google have also published findings on similar attacks against their own models, suggesting a problem that extends well beyond a single company.
A Tense Policy Backdrop
About two months before CNBC confirmed the letter, the White House Office of Science and Technology Policy issued a memo promising to help AI companies detect industrial-scale distillation attacks and coordinate their response. In April, Donald Trump had accused China of stealing AI “at industrial scale.” Anthropic argues that Alibaba “ignored warnings from the Trump administration” by continuing this kind of extraction.
The timing overlaps with a separate dispute between Anthropic and the administration over access to its most advanced models. Anthropic said that earlier in June it received an export-control directive ordering it to suspend access to its latest models, Fable 5 and Mythos 5, for any foreign national, including its own foreign-national employees. According to the company, the administration cited national-security authorities without specifying its concern. Both sides say they are working toward a quick resolution, but no availability date has been announced. The sources also use the name “Mythos Preview” without explaining how it relates to “Mythos 5.”
Obfuscation, Proxies and National Security
Anthropic says the operators used “obfuscation techniques and proxy networks” to avoid detection. It warns of a growing “circumvention economy,” fed by Chinese demand for reliable obfuscation tools, that could fuel a wider network of future distillation attacks. The letter calls these attacks “widespread” and says they turn “hundreds of billions of dollars of American investment and R&D into a massive subsidy for our geopolitical competitors.”
Anthropic also links the attacks to U.S. national-security concerns. It cites U.S. Department of Defense allegations that Alibaba, as well as BYD and Baidu, have ties to the Chinese military. Those companies deny the claims. The BBC reports that Alibaba has sued the U.S. government to have its name removed from the Pentagon blacklist. In its response, Alibaba states:
“Alibaba is governed by an independent board, none of whom has any military affiliation. Its products and services are built for retail, logistics, and enterprise information technology — not weapons, defense, or intelligence.”
Alibaba, statement cited by the BBC
Anthropic does not consider that response conclusive. It warns that without stronger intervention, the attacks will help China reach Mythos Preview-level capabilities sooner.
Three Legislative Levers
Anthropic is urging Congress to penalize companies that conduct such attacks and to strengthen safeguards against the theft of U.S. technology. It proposes three objectives:
- Update antitrust law so AI companies can share information about evolving Chinese tactics.
- Tighten export controls on chips and advanced computing to limit Chinese access to the infrastructure needed for training.
- Establish penalties that make it harder and costlier for Chinese labs to rely on distillation.
The company also suggests restricting Chinese firms’ access to U.S. models and advanced U.S. chips, as well as their reliance on data centers outside China.
“We believe combating the threat of illicit distillation requires coordinated action between government and industry, and we will continue working with Congress and the Administration to maintain American AI leadership.”
Spokesperson, Anthropic
Outlook and Scenarios
The figures published by Anthropic remain allegations. Alibaba’s response rejects them in substance, and no signed admission has surfaced. Alibaba is listed on the New York Stock Exchange, and the BBC reports that Anthropic, alongside OpenAI, is preparing a major stock-market debut that could make it one of the world’s most valuable companies. Beyond these market-status details, the sources report no quantified share-price reaction to the accusations.
Three scenarios stand out. In the first, export controls and sector cooperation meaningfully reduce the effectiveness of distillation. In the second, pressure remains mostly rhetorical and attackers shift toward more sophisticated proxy networks. In the third, tougher U.S. measures fragment the ecosystem, with Chinese models trained on separate infrastructure. In the near term, the outcome will depend largely on the legislative timetable and on how the dispute over access to Fable 5 and Mythos 5 is resolved.
Sources
Cet article est publié à titre informatif et éducatif. Il ne constitue en aucun cas un conseil en investissement. Faites vos propres recherches (DYOR) avant toute décision.
