Anthropic Accuses Alibaba of Illicit Large-Scale Claude AI Model Distillation

Anthropic has formally accused Chinese tech giant Alibaba of a coordinated, illicit campaign to extract proprietary capabilities from its Claude AI model. The AI developer alleges that operators linked to Alibaba utilized thousands of fraudulent accounts to perform nearly 29 million queries—specifically identified as 28.8 million exchanges—to map the reasoning pathways and structural parameters of Claude. Anthropic describes this activity as a form of "distillation," where malicious actors analyze a large model’s responses to replicate its advanced behaviors in weaker models, a direct violation of their terms of service.

The conflict highlights escalating tensions in the global AI race, where intellectual property and model architecture are increasingly treated as strategic national and corporate assets. Anthropic claims the campaign bypassed existing safety protocols to enable the unauthorized replication of its proprietary 'Mythos Preview' performance. This dispute has been escalated to U.S. legislators, including Senator Elizabeth Warren and Senator Tim Scott, as the industry grapples with the threat of international firms cloning foundation models to compete with Western AI labs.

In response, Anthropic has implemented aggressive technical defenses, including enhanced anomaly detection and stricter rate-limiting protocols. The incident has intensified bipartisan calls for stricter AI security standards and cross-border regulatory frameworks. As companies like Anthropic prepare for major milestones such as initial public offerings, this breach underscores the vulnerability of frontier models. The industry expects a significant shift toward more rigorous API authentication, stricter oversight of model interaction data, and potential legal consequences for entities involved in large-scale cross-border data harvesting.