Anthropic accuses Alibaba of largest known AI distillation attack: 28.8 million Claude queries through 25,000 fake accounts
What it really says
On June 10, 2026, Anthropic sent a letter to US Senate Banking Committee Chair Tim Scott and Ranking Member Elizabeth Warren alleging that operators affiliated with Alibaba's Qwen AI lab conducted the largest known distillation attack on its Claude models. According to the letter, approximately 25,000 fraudulent accounts generated 28.8 million exchanges with Claude between April 22 and June 5, 2026, a period of 44 days. The method used is called adversarial distillation: systematically querying a superior AI model with carefully crafted prompts to extract its reasoning patterns and knowledge structures, then using the collected responses to train a cheaper rival model. The campaign specifically targeted Claude's software engineering, agentic reasoning, and cybersecurity capabilities, which are the most commercially valuable aspects of the model, particularly those embodied in the Mythos Preview model. This is not the first such accusation from Anthropic: in February 2026, the company accused Chinese AI startups DeepSeek, MiniMax, and Moonshot AI of conducting 16 million exchanges through about 24,000 fake accounts. US Senators Bill Hagerty and Andy Kim are now working on legislation that would enable sanctions against entities conducting such distillation campaigns.
Our assessment
This story merits a yellow rating because it highlights a real security concern in the AI industry while important context must be considered. The legitimate concern: if AI capabilities can be systematically extracted through automated queries at scale, it undermines research and development investments and creates incentives for a form of digital industrial espionage. The fact that Anthropic is raising such allegations for the second time in a few months suggests a systemic problem. At the same time, important caveats exist: Alibaba denies the allegations, and the figures come exclusively from Anthropic and have not been independently verified. Distillation is also not the same as stealing source code or model weights. Distilled models typically do not reach the full capability of the original. Furthermore, Anthropic has a commercial interest in making these allegations public, as they strengthen the case for stricter export controls and technical protections. The political response in the US, however, indicates that the allegations are being taken seriously.
Relevance for Germany
This development has several dimensions for Germany and Europe. First, the case demonstrates how vulnerable even the most advanced AI models are to systematic exploitation, a problem that also affects European companies relying on Claude or other US models. Second, the incident further escalates geopolitical tensions between the US and China in the AI sector, which can impact European companies maintaining business relationships with both sides. Third, the case raises fundamental questions about intellectual property protection for AI models that are also relevant to the EU AI Act: how do you protect AI capabilities when model weights are not stolen, but only their outputs are systematically copied? Fourth, the incident underscores Europe's dependence on US AI infrastructure: if Chinese actors successfully distill US models, the resulting models could be offered as cheaper alternatives on the European market with unclear provenance of the underlying capabilities.
Fact check
The allegations originate from a letter by Anthropic to US senators, whose contents are consistently reported by CNBC, Bloomberg, and The Next Web. The specific figures of 25,000 accounts and 28.8 million queries are identically cited across all reports but come exclusively from Anthropic's own account. According to CNBC, Alibaba either declined to comment or denied the allegations. The earlier accusations against DeepSeek and other Chinese startups from February 2026 are documented through Bloomberg reports. The legislative response from Senators Hagerty and Kim is confirmed by multiple sources. No independent verification of Anthropic's figures exists to date.
Source
- • https://www.cnbc.com/2026/06/24/anthropic-alibaba-distillation-campaign.html
- • https://thenextweb.com/news/anthropic-accuses-alibaba-distillation-claude-qwen
- • https://www.tweaktown.com/news/112361/anthropic-says-alibaba-used-25000-fake-accounts-to-query-claude-28-8-million-times-and-train-qwen-off-the-results/index.html
- • https://runtimewire.com/article/anthropic-alibaba-qwen-claude-distillation-claims