Anthropic accuses Alibaba of ‘brazen’ bid to copy Claude

TL;DR:

• Anthropic has told the US Senate Banking Committee that Alibaba ran the “largest campaign to illicitly extract Claude’s capabilities”.
• It alleges 25,000 fraudulent accounts generated 28.8m exchanges between late April and early June, targeting agentic reasoning and software engineering.
• The accusation extends a pattern Anthropic previously pinned on DeepSeek, Moonshot and MiniMax.

The frontier AI rivalry between Washington and Beijing has a new flashpoint. In a 10 June letter to Congress, seen by both Reuters and the Financial Times, Anthropic accused Chinese commerce giant Alibaba of illicitly accessing its Claude model through tens of thousands of fake accounts — the largest such campaign the company says it has detected.

A familiar playbook, a bigger scale

Anthropic alleges the operation ran between late April and early June, using around 25,000 fraudulent accounts to generate more than 28.8m exchanges with Claude. The targets, it said, were Claude’s “most valuable capabilities, such as agentic reasoning, software engineering, and long-horizon tasks”. The technique — “distillation” — trains a weaker model on a stronger one’s outputs, and Anthropic has levelled similar claims before: in February it named DeepSeek, Moonshot and MiniMax, whose combined activity it now puts at 16m exchanges across 24,000 accounts.

The FT reported that Anthropic urged Congress to “close loopholes” on Chinese access to advanced US chips and to penalise labs behind distillation attacks, describing Alibaba as having acted “brazenly” even after a White House memo on industrial-scale IP theft. Alibaba, which this week asked a US court to remove it from a Pentagon blacklist of firms with alleged military ties, did not respond to requests for comment.

For UK readers, Anthropic is no peripheral player. The company sits at the centre of Britain’s AI-policy debates, from the export-control saga around its Mythos model to shifting US security assessments. The accusation also arrives as Chinese cybersecurity firm 360 unveils its own answer to Anthropic’s Mythos, underlining how quickly the capability contest is escalating.

Looking forward

The letter lands days before a Senate hearing on AI, and it sharpens the case for tighter controls just as US firms weigh how exposed their models are. For British businesses building on frontier models, the episode is a reminder that the tools they rent sit inside a geopolitical contest whose rules — on access, security and IP — are still being written.