A new high-stakes controversy has erupted across the global artificial intelligence sector, as leading United States AI developer Anthropic has leveled serious allegations against Chinese tech and e-commerce giant Alibaba, claiming the firm orchestrated an unprecedented large-scale operation to steal the proprietary capabilities of Anthropic’s flagship Claude AI model.
In a formal June 10 letter addressed to top U.S. senators Tim Scott and Elizabeth Warren — a copy of which has been reviewed by the BBC — the San Francisco-based AI firm claims that actors tied to Alibaba carried out nearly 29 million interactions with Claude through a network of thousands of fraudulently created accounts. Anthropic frames this activity as the largest known operation of its kind to date, conducted via a technique called a “distillation attack.” This method involves pulling large volumes of output from a sophisticated, well-trained large language model to teach a smaller, competing model to replicate its performance at a fraction of the original development cost.
According to the allegations, the operators specifically targeted Claude’s most commercially and technically valuable functionalities: its advanced capacity to process long-form, complex prompts and its nuanced decision-making frameworks. Anthropic argues that this kind of industrial-scale attack allows foreign competitors to repurpose hundreds of billions of dollars in U.S. private sector research and development investment as a de facto subsidy for their own tech advancement, creating unfair competitive advantages and threatening U.S. national technological leadership.
The letter also expands the scope of the accusations beyond Alibaba, linking other major Chinese firms including automaker BYD and search giant Baidu to similar activities, and repeating longstanding unproven claims that these companies have ties to the Chinese military that pose risks to U.S. defense interests. All the named firms have repeatedly denied these assertions in the past, and Alibaba recently took legal action against the U.S. government to challenge its inclusion on a Pentagon blacklist over the unsubstantiated connections. The BBC has reached out to both Alibaba for a response to the new allegations and Anthropic for additional supporting evidence, but has not received comment as of yet.
This is not the first time a leading U.S. AI firm has levied such claims against Chinese entities. OpenAI, Anthropic’s primary competitor and creator of the ChatGPT platform, previously made identical accusations about distillation attacks carried out by Chinese-linked groups.
The controversy emerges as Anthropic, alongside OpenAI, prepares for a highly anticipated initial public offering that market analysts expect to value both companies among the world’s most valuable technology firms. Notably, Anthropic itself has faced prior scrutiny over cybersecurity risks: some of its more advanced, unreleased models have raised red flags among security experts for their demonstrated ability to identify and exploit vulnerabilities in computer systems.
Anthropic is now pushing U.S. lawmakers to introduce harsh new penalties for entities found to carry out these kinds of IP theft operations, and to strengthen federal measures designed to block the misappropriation of U.S.-developed artificial intelligence technology.