Anthropic: Alibaba Accessed Claude 28.8M Times
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Anthropic told US officials that ~25,000 accounts accessed Claude 28.8 million times between April and June, calling the activity a large-scale "adversarial distillation" campaign
- Coverage from Bloomberg, Reuters, WSJ, FT, BBC, and CNBC uniformly framed the effort as "brazen" and "illicit," with Nikkei Asia calling it the "largest known distillation attack" on Claude
- Anthropic alleged the scraping was directed at improving Alibaba's Qwen model, per RuntimeWire's framing that the accounts were used to "mine Claude for Qwen"
- The company's letter went to US officials including Senators Elizabeth Warren and Tim Scott, and Anthropic is seeking tougher US export curbs on Chinese AI labs
- The accusation lands ahead of expected new US restrictions on the newest AI models, per Türkiye Today's framing of the timing as occurring "before curbs on newest models"
Why it matters: Anthropic's 28.8M-interaction allegation gives US regulators a specific, quantified case to justify tightening curbs on Chinese AI labs — and the accusation lands right as Washington weighs new model restrictions, per Türkiye Today's note that it came "before curbs on newest models." For Alibaba, the public framing as the operator of 25,000 fake accounts puts Qwen's training provenance under direct public and regulatory scrutiny.
