Anthropic lodged a formal complaint with the Senate Banking Committee, accusing Alibaba of running the “largest campaign to illicitly extract” its Claude model. The Chinese e‑commerce giant allegedly created 25,000 fake accounts that generated more than 28.8 million interactions with Claude between late April and early June, breaching the service’s terms and potentially gave Chinese developers a shortcut to replicate cutting‑edge AI.

Anthropic’s letter warned that the breach targeted Claude’s most valuable capabilities—agentic reasoning, software‑engineering prompts and long‑horizon tasks—capabilities also embedded in its Mythos model used in cyber‑offensive programs. The firm urged Congress to close loopholes that let Chinese AI labs access U.S. chips and to penalise entities responsible for “distillation attacks.” Such illicit training could accelerate Chinese competitors’ ability to field models rivaling U.S. offerings.

The accusation arrives as Alibaba fights removal from a Pentagon blacklist that labels it a People’s Liberation Army proxy. A recent White House memo, based on declassified intelligence, alleged the Hangzhou‑based group supplied technology for undisclosed military operations, a claim Alibaba denies. The dispute underscores rising geopolitical risk for AI vendors and could tighten export controls on advanced models. Investors monitor fallout for cloud contracts.