Anthropic has gone on the record with detailed accounts of what it describes as systematic model distillation campaigns carried out by three Chinese AI companies: Alibaba, Moonshot AI, and DeepSeek. The disclosure, reported by TechCrunch, represents one of the most specific public accusations a major AI lab has leveled against competitors over the practice of using a commercial model's outputs to train a rival system.
What Distillation Means and Why It Matters
Model distillation, in this context, refers to querying a frontier AI model at high volume and using those responses as training data for a separate model. It is technically possible in many commercial settings but generally prohibited by terms of service. When done covertly and at scale, it can allow a developer to absorb the capabilities of an expensive, heavily trained model without bearing the underlying research costs. The practice sits in a legal and ethical gray zone, and Anthropic's decision to name specific companies publicly signals it intends to treat these incidents as something more serious than routine terms-of-service violations.
The Alibaba case has been developing for some time. Earlier reporting established that Alibaba used roughly 25,000 fake accounts to extract Claude AI data, a scale that suggests a structured, deliberate operation rather than opportunistic scraping. Anthropic's latest disclosure appears to add further detail to that picture, placing it alongside the Moonshot AI and DeepSeek incidents as part of a broader pattern.
Key Facts
- Anthropic publicly named Alibaba, Moonshot AI, and DeepSeek as having conducted distillation campaigns against Claude.
- The Alibaba operation allegedly involved approximately 25,000 fake accounts used to systematically query Claude.
- Moonshot AI and DeepSeek are accused of rerouting user traffic through Claude without disclosure.
- Anthropic's terms of service prohibit using Claude's outputs to train competing models.
- The disclosures mark a rare instance of an AI lab publicly detailing alleged IP extraction by named competitors.
The Moonshot AI and DeepSeek situations involve a somewhat different mechanism. According to Anthropic, those companies allegedly diverted user requests to Claude rather than serving them with their own models, effectively passing off Claude's responses as their own. Anthropic's claims that Moonshot and DeepSeek rerouted users to Claude point to an approach that goes beyond data harvesting and into questions about product misrepresentation.
The scale and coordination of these campaigns suggest they were not accidental policy oversights but deliberate strategies to absorb Claude's capabilities at a fraction of the development cost.Anthropic, via TechCrunch
Fallout and the Broader Context
The accusations have already produced reactions on the Chinese side. Alibaba moved to distance itself by banning Claude Code internally, a decision that came with its own controversy after some reports suggested Alibaba cited concerns about Anthropic software behavior as justification. The response underscores how tangled these disputes can become when geopolitics, competitive pressure, and genuine security concerns intersect in the same narrative.
For Anthropic, going public with named companies carries real stakes. It invites legal scrutiny, diplomatic friction, and the possibility of retaliation. At the same time, the lab appears to be calculating that transparency serves its interests better than quiet enforcement. Claude is Anthropic's primary commercial product, and the integrity of its training pipeline is central to the value it offers enterprise customers.
The incidents also feed into a wider conversation about how Western AI labs protect their models in a competitive global market. Distillation attacks are not unique to Chinese companies, but the scale alleged here and the involvement of firms with significant state-adjacent relationships adds a dimension that goes beyond ordinary competitive intelligence. Whether Anthropic pursues legal remedies or relies on technical countermeasures, the disclosures have put the issue firmly on the industry's agenda. Those following the latest Claude AI news will find this episode hard to ignore as a signal of what frontier AI competition increasingly looks like.
