In brief
Anthropic has accused Alibaba of extracting the capabilities of its Claude AI model through massive distillation attacks, claiming nearly 29 million illicit exchanges. The San‑Francisco company sent a letter to Senators Tim Scott and Elizabeth Warren outlining the alleged campaign and warning of national security risks. Alibaba denies the allegations and recently sued to remove its name from a Pentagon blacklist. Both Anthropic and competitors like OpenAI face scrutiny over similar accusations. The story underscores heightened tensions in the global AI arms race.
In a 10 June letter, Anthropic—an AI firm behind the Claude model—accused Alibaba of carrying out “the largest campaign to illicitly extract Claude’s capabilities.” The U.S. company says Alibaba‑linked operators conducted almost 29 million exchanges with Claude, using thousands of fraudulent accounts.
Anthropic’s allegations hinge on “distillation attacks,” where answers from a stronger model are copied to train a weaker one. It claims Alibaba targeted Claude’s advanced features such as handling long, complex tasks and decision‑making, enabling the Chinese competitor to re‑package U.S. technology as its own.
The report also warns that such industrial‑scale attacks could threaten U.S. military security, describing the theft as a “massive subsidy” for China’s geopolitical rivals.
Alibaba has denied all allegations and, last week, sued the U.S. government to have its name removed from a Pentagon blacklist. OpenAI has faced similar claims, and cybersecurity experts cite concerns over Anthropic’s Mythos model.
Anthropic is preparing for a high‑profile IPO, potentially becoming one of the world’s most valuable AI companies, but the allegations add a complex layer to its market debut.
For further context, see related stories on China’s AI developments and the recent Chinese AI apps reshaping the landscape.
