Anthropic accuses Alibaba of stealing Claude AI capabilities to train rival models

U.S. AI company Anthropic has accused Chinese technology giant Alibaba of orchestrating the largest known illegal extraction of its Claude AI model capabilities, a campaign Anthropic describes as a brazen "distillation attack" designed to train Alibaba's own competitive models. The allegation details that operators linked to Alibaba and its AI lab Quen utilized nearly 25,000 fraudulent accounts to generate over 28.8 million interactions with Claude, specifically targeting its advanced reasoning and decision-making processes to effectively copy its abilities. This incident, which Anthropic reported to U.S. lawmakers in a letter urging legislative penalties for such theft of American technology, represents a significant industrial-scale effort by Chinese firms to harvest and rebrand advanced U.S. AI advancements as their own. The dispute highlights growing tensions over AI intellectual property and data security, as it raises concerns about the illicit use of fraudulent networks to bypass access controls and steal proprietary model functionalities.