In brief
Artificial intelligence giant Anthropic has accused Alibaba of orchestrating the largest illicit extraction of its Claude model’s capabilities, claiming the use of nearly 29 million fake accounts to drain the system’s knowledge. The firm’s letter to Senators Tim Scott and Elizabeth Warren demands sanctions against the actors and urges Congress to bolster safeguards for U.S. tech exports. Alibaba, which has denied the allegations and sued the U.S. government to remove its name from a defense blacklist, has tied its success to these claims amid growing tensions over AI safety. The incident highlights the broader clash between Western AI developers and Chinese firms accused of distillation attacks, a practice that could undermine U.S. competitive edge.
The U.S. artificial‑intelligence powerhouse Anthropic has taken aim at Chinese e‑commerce titan Alibaba, accusing it of a "brazen" and "illicit" operation that siphoned the company’s Claude model’s capabilities through thousands of fake accounts.
In a letter dated 10 June, Anthropic warned that the campaign involved almost 29 million exchanges with Claude, describing it as the largest extraction effort known. The firm argued that the actors behind Alibaba leveraged distillation attacks—extractions of a powerful AI’s outputs to train a weaker one—to steal the model’s advanced reasoning and decision‑making skills.
Anthropic urged Senators Tim Scott and Elizabeth Warren to penalize the companies behind such attacks and to intensify U.S. measures against technology theft. It also cited potential threats to national security, noting that similar tactics have enabled Chinese competitors to repackage U.S. AI innovations as their own.
Alibaba and other implicated firms have denied the allegations. In a recent day, Alibaba filed a lawsuit against the U.S. government seeking removal from the Pentagon blacklist, a move that underscores its push against perceived unfair restrictions.
Other U.S. developers, including OpenAI, have raised similar concerns over Chinese distillation practices that seem to undermine the cost advantage of American AI research. Anthropic’s name is also ring‑fenced for an upcoming highly‑anticipated public listing that could catapult it to near‑mythic market valuations.
With advanced models such as Mythos tripping over cybersecurity failings—particularly their propensity to exploit system weaknesses—industry experts warn that the clash may intensify in the coming months as governments tighten export controls. The unfolding saga is being watched keenly by policymakers, investors and AI researchers alike.
