US artificial intelligence developer Anthropic has formally accused Chinese e-commerce and technology conglomerate Alibaba of 'brazenly' and 'illicitly' extracting the capabilities of its Claude AI model. The serious allegations were detailed in a letter sent to two members of the US Congress, Senators Tim Scott and Elizabeth Warren, on 10th June.
According to Anthropic, operators linked to Alibaba engaged in what it described as the 'largest extraction campaign of its kind', involving almost 29 million interactions with Claude. These interactions were reportedly conducted using thousands of fraudulent accounts. The San Francisco-based company claims that these actions constitute 'distillation attacks', a method where responses from a more powerful AI model are extracted to train a weaker one.
Anthropic stated that the Alibaba-linked operators specifically targeted Claude's most valuable functionalities, including its capacity for handling complex and lengthy tasks, as well as its sophisticated decision-making processes. The AI developer argued that such attacks are carried out on an 'industrial scale', enabling Chinese companies to acquire and rebrand US AI capabilities as their own, significantly reducing their research and development costs.
The letter further suggested that these alleged attacks pose a threat to US military interests. Anthropic cited claims from the US Department of Defense that link Alibaba, alongside other major firms such as car manufacturer BYD and tech company Baidu, to the Chinese military. Anthropic asserted that 'distillation attacks turn hundreds of billions of dollars in American investment and [research and development] into a massive subsidy for our geopolitical competitors'.
While Alibaba and the other named companies have consistently denied any ties to the Chinese military, Alibaba recently initiated legal action against the US government in an effort to have its name removed from the Pentagon's blacklist. This incident is not isolated, as other US AI developers, including OpenAI, have previously accused Chinese competitors of employing similar distillation tactics to train their AI models to rival American technology at a fraction of the cost.
Anthropic, a prominent AI developer alongside OpenAI, is currently preparing for a potential stock market debut, which could position it among the world's most valuable companies. However, some of its more advanced models, such as Mythos, have also raised cybersecurity concerns due to their potential ability to identify vulnerabilities in computer systems.