Anthropic accuses Alibaba of largest known AI theft - 28.8 million queries in six weeks
What it really says
In a letter to US senators and the White House, Anthropic has accused Chinese technology conglomerate Alibaba of executing the largest documented campaign to steal AI capabilities. According to Anthropic's testimony before the US Senate, operators linked to Alibaba's Qwen laboratory conducted approximately 28.8 million unauthorized exchanges with Claude over a six-week period. They employed roughly 25,000 fraudulent accounts and commercial proxy services to circumvent geographic access restrictions. The method used is called adversarial distillation: systematically querying an advanced AI model with targeted prompts to extract its reasoning patterns and data structures. Competitors can use these extracted patterns to train their own models, bypassing millions of dollars in research and development costs. The US Department of Commerce had already issued an export control directive on June 12, 2026, ordering Anthropic to suspend access to Fable 5 and Mythos 5 for all foreign nationals worldwide. The new accusations now provide a possible explanation for this drastic measure.
Our assessment
This story warrants a yellow rating because it reveals a real security risk while also demonstrating that protective measures work. The legitimate concern: if state-backed actors can systematically spy on AI models, it calls into question the integrity of the entire AI ecosystem. 28.8 million queries through 25,000 fake accounts demonstrate an industrial scale far beyond ordinary terms-of-service violations. However, there are also reassuring aspects: Anthropic detected and documented the campaign, showing that monitoring systems work. The US government also responded - albeit with a measure that affected all international users, not just the alleged attackers. What must be viewed critically, though, is the proportionality: the export controls that followed this situation also denied European users access to Fable 5 and Mythos 5, even though Europe was not the source of the threat. This raises the question of whether uninvolved third parties are being punished for geopolitical conflicts.
Relevance for Germany
This story is relevant to Germany for several reasons. First, German users and companies were directly affected by the export controls issued in response to this campaign - they lost access to Anthropic's most capable models despite Germany not being a party to this conflict. Second, the case demonstrates the vulnerability of AI systems that German companies increasingly rely on: if even the most advanced models can be copied through systematic distillation, this also concerns the protection of trade secrets processed via AI interfaces. Third, the episode underscores the necessity of European AI sovereignty: as long as Europe lacks its own frontier models, it remains a bystander in a geopolitical conflict whose consequences it directly feels. Fourth, the accusations could accelerate the trusted-partners framework negotiations begun at the G7 summit in Evian - though under tougher conditions.
Fact check
The allegations come from an official letter by Anthropic to the US Senate and the White House, which became public on June 24, 2026. The specific figures - 28.8 million queries, 25,000 fraudulent accounts, six-week campaign duration - are consistently reported by CNBC, TechTimes, and BusinessToday, based on Anthropic's own account. It is important to note that these are Anthropic's accusations - no independent verification or admission from Alibaba exists. Alibaba's Qwen laboratory has not publicly commented on the allegations. The temporal connection to the export control directive of June 12, 2026 is drawn in reporting but is not officially confirmed as a causal link.
Source
- • https://www.cnbc.com/2026/06/24/anthropic-alibaba-distillation-campaign.html
- • https://www.techtimes.com/articles/319105/20260625/alibaba-ran-largest-known-ai-theft-campaign-against-claude-anthropic-tells-senate.htm
- • https://www.businesstoday.in/technology/artificial-intelligence/story/anthropic-accuses-alibaba-of-largest-ever-attempt-to-extract-claude-ai-capabilities-539087-2026-06-25
- • https://creati.ai/ai-news/2026-06-25/anthropic-accuses-alibaba-illicitly-extracting-claude-ai-model-capabilities/