Recently, the U.S.-based AI company Anthropic, in a letter addressed to U.S. senators, accused Alibaba of unlawfully obtaining capabilities from its Claude model via a 'distillation attack' to train its own smaller AI systems. Anthropic labeled this as one of the most significant corporate espionage incidents to date, one that could potentially jeopardize the technological security of the United States and its allies. According to publicly available information, Anthropic claimed that between April 22 and June 5, 2026, Alibaba interacted with the Claude model over 28.8 million times through nearly 25,000 fraudulent accounts to extract its model capabilities. However, this accusation has not yet been substantiated by judicial or regulatory bodies and remains a subject of intense debate. On one hand, distillation technology is a widely accepted practice in the AI industry. It involves training smaller models using the outputs of larger models to facilitate knowledge transfer and model compression. Leading AI labs globally openly acknowledge conducting systematic evaluations and capability comparisons of competing models. Equating normal model usage with 'theft' is seen by many as a deliberate misrepresentation of industry standards. On the other hand, Anthropic itself has encountered data compliance issues. The company was compelled to pay a $1.5 billion settlement for illegally downloading over 7 million copyrighted books from pirated websites to train its models. This raises questions about whether Anthropic's accusations against the Chinese company could be seen as a case of 'the pot calling the kettle black.' Additionally, some observers argue that Anthropic's accusations are part of a premeditated effort by the U.S. government and corporations to stifle China's AI industry through political pressure and smear campaigns, with the aim of curbing the growth of Chinese AI firms.