In a bold move that highlights the growing conflict in the artificial intelligence arms race, Anthropic, an American AI firm, has formally leveled allegations of the largest known attack to date against its AI systems by the Chinese technology giant, Alibaba. According to the allegations in a confidential letter sent to key members of the United States Senate Banking Committee, Alibaba engaged in a “massive and systematic plunder” of capabilities from Anthropic’s Claude AI platform, which involved nearly 25,000 fraudulent accounts and more than 28.8 million interactions with Anthropic’s system. The discovery comes amidst a growing geopolitical intrigue in how AI is being developed, as governments and businesses alike worry about the impact of potential IP infringement and the need to stay ahead of the technology curve.
The letter signed by the company on June 10, sent to Sen. Tim Scott and Sen. Elizabeth Warren, describes an elaborate scheme, allegedly carried out by operators linked to Alibaba and its AI research unit, Alibaba Qwen. The campaign ran from April 22 to June 5, 2026, and the company claims it took place to try to duplicate Claude‘s high-level abilities by a process called distillation. The idea behind this is to train a less advanced model using the results from the more powerful one, so that the more powerful model benefits the more advanced model. Anthropic has described it as a “strategic operation” to help China meet the performance of its advanced Mythos Preview features, which suggests it may be more than just random occurrences.
Model distillation has become a controversial topic in the AI sector, giving companies an edge over investing in advanced AI models when the expense and expertise of creating AI models are significant. Knowledge mining from existing models such as Anthropic’s Claude can lead companies to similar performance levels, but with much smaller investments in computing power and specialized personnel. Like the case against Chinese AI company DeepSeek, earlier this year, which was also accused of infringing upon the patent, Anthropic is alleging that Alibaba’s model infringes upon a patent held by its chief rival, Google.Like the case against Chinese AI company DeepSeek earlier this year which shook up the technology industry with its low-cost model, Anthropic is claiming that Alibaba’s model violates a patent held by its primary competitor, Google. In that period, Anthropic found more than 150,000 exchanges in DeepSeek’s campaigns, and more than 3.4 million and 13 million exchanges for Moonshot AI and MiniMax respectively. These campaigns were becoming more sophisticated and intense, the company said, and tackling the threat would take swift, concerted effort by industry, policy makers and the global AI community.
This comes amid growing worries about the Chinese access to U.S. AI technology. In April, the White House publicly alleged that China is stealing the U.S. AI labs’ IP rights on an industrial scale, calling it a “threat to U.S. economic and national security.” The Pentagon is also acting, with Alibaba being placed on its list of Chinese military companies this month, which the company is contesting. Meanwhile, Commerce Department has reportedly been waiting on a trade blacklist for DeepSeek, while officials leave an interagency governmental committee assessment of it as a national security risk to stand, hoping to keep tensions at bay with Beijing.
Anthropic’s letter is notable for its timing, as other regulatory developments around the use of advanced AI models unfold. The Commerce Department has imposed controversial restrictions on Anthropic’s newest AI models, Mythos and Fable, just two days after the company sent its letter to the Senate committee.Two days after sending its letter to the Senate committee, the Commerce Department has slapped controversial restrictions on Anthropic’s latest AI models, Mythos and Fable. Officials said they were worried the systems might be used by the military intelligence services of countries of concern, such as China, and that Anthropic could expand its prohibition of access to the models to the entire world. The move underscores the intricate nature of the challenges posed by AI firms concerning the intersection of commercial interests and national security issues, especially as their technology grows more sophisticated and significant.
Anthropic has voiced its backing for U.S. government initiatives to fight this type of attack, such as collaborating with AI private sector business partners in an effort to share threat intelligence and other collaborative measures. The openness of the company to reveal the campaign against Alibaba reveals a strategic shift towards communicating with policy makers and emphasizing the extent of the problem facing the US AI industry. But the silence of Alibaba on requests for comment puts the allegations in a limbo and the larger ramifications of U.S.-China technology relations are open to interpretation.
The charges brought against Alibaba speak to the underlying question of the viability of global collaboration on the development and research of AI. Although open scientific exchange has been a feature of the industry in the past, the strategic value of AI brings a growing emphasis on safeguarding proprietary technology. This conflict between humans and machines will continue to grow more fierce as AI systems become more powerful and their scope of uses in civilian and military applications also increases. Anthropic’s distillation campaigns are a form of threat that is especially difficult to detect and prevent because it involves technology transfer that takes place in a “gray space” of technology transfer, which is not easily captured by traditional legal or regulatory frameworks.
