Anthropic alleges illicit Claude access by Alibaba

Anthropic PBC has alleged that Alibaba Group Holding Ltd. (9988.HK) mounted a large-scale effort to illicitly access its Claude artificial intelligence model. The AI developer says the activity directly challenged its choice to keep Claude’s services out of China, prompting a formal warning to US policymakers. The accusations center on attempts to tap some of Claude’s most advanced capabilities rather than routine or limited use.

In materials shared with several US senators and White House officials, Anthropic describes what it views as an organized campaign tied to Alibaba’s (9988.HK) AI operations. The company characterizes the effort as the biggest known attempt so far by a Chinese firm to piggyback on the work of top US AI labs. The allegations focus on the scale, coordination and technical focus of the activity rather than on isolated misuse by individual users.

Anthropic states that the attempt to reach Claude’s systems relied on thousands of fraudulent accounts. These accounts were described as being used to access the model in ways that bypassed normal usage patterns and controls. By distributing activity across many accounts, the operators allegedly sought to avoid detection while systematically querying Claude.

The company links these accounts to operators associated with Alibaba’s Qwen AI lab. While the detailed technical methods are not disclosed in the available reports, Anthropic presents the scale of the account network as central to its concerns. The breadth of the alleged account activity underpins Anthropic’s description of the campaign as unprecedented in size for a Chinese technology company targeting US AI models.

Anthropic says the operators focused on Claude’s “most prized” capabilities, including software engineering functions and agentic reasoning. These features enable complex code generation and advanced reasoning workflows, which are considered among Claude’s key differentiators. The pattern of requests described by Anthropic suggests a sustained effort to draw out high-value capabilities rather than general-purpose usage.

By emphasizing these targets, Anthropic frames the incident as more than simple unauthorized access. The company suggests the activity aimed at systematically extracting sophisticated behaviors and know-how from Claude. This focus reinforces Anthropic’s view that the campaign was intended to benefit an external AI development program rather than casual users.

Anthropic delivered its allegations in a letter addressed to multiple US senators and officials at the White House. The outreach indicates that the company is seeking government attention on what it regards as a significant cross-border AI and intellectual property issue. The letter presents the incident as a case study in how advanced AI models can be targeted at scale.

The reports describing the letter do not detail any specific regulatory steps that may follow, nor do they include a public response from Alibaba to the claims. Nonetheless, the episode highlights increasing scrutiny of how frontier AI models are accessed and used across jurisdictions. It also illustrates how commercial disputes over cutting-edge AI may intersect with national policy discussions on technology security and competition.