Anthropic: Alibaba Hit Claude 28.8M Times via 25K Accounts
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Anthropic accused Alibaba of a large-scale adversarial distillation campaign, alleging Alibaba-linked operators accessed Claude 28.8 million times from April to June 2026 through approximately 25,000 fraudulent accounts, per Bloomberg and Reuters.
- Anthropic detailed the allegations in a letter to US officials including Senators Elizabeth Warren and Tim Scott (per Benzinga), claiming the extracted data was used to train Alibaba's Qwen models.
- Alibaba shares dropped 4.43% on the news, with at least 18 outlets — spanning US, UK, Chinese, Indian, Japanese, and Turkish media — uniformly framing the episode as the 'largest known' or 'largest-ever' distillation attack on a frontier AI model.
- Anthropic timed the disclosure ahead of expected curbs on its newest models, per Türkiye Today, and is now explicitly asking Washington to restrict Chinese AI labs, per Fortune India and PYMNTS — converting a corporate IP complaint into a policy ask.
- The scale of the alleged operation — 28.8 million exchanges across ~25,000 accounts over roughly two months — far exceeds typical API abuse, with Cyber Security News and Implicator.ai emphasizing the use of fake accounts to mask the traffic.
- Alibaba has not been shown to publicly respond in the source coverage, even as outlets from Nikkei Asia to DigiTimes framed the dispute as a new front in the US-China AI rivalry.
Why it matters: Anthropic's letter does double duty: it publicizes a 28.8 million-access alleged IP theft while explicitly requesting tougher US export curbs on Chinese AI labs, meaning a single corporate complaint is now being leveraged into regulatory action. Alibaba's 4.43% stock drop shows investors are pricing in both reputational damage and the prospect of new US restrictions on its Qwen roadmap.
