EVERYTHING
According to Thenextweb, Anthropic has identified Alibaba as the source of what it describes as the largest distillation campaign ever launched against a U.S. artificial intelligence company. The allegations center on a coordinated effort to harvest Claude's capabilities at an industrial scale to train cheaper rival systems.
Scale and scope of the distillation attack
The letter sent to government officials detailed a sophisticated operation where actors linked to Alibaba’s Qwen lab utilized nearly 25,000 fake accounts. These accounts were used to facilitate roughly 29 million exchanges with Claude during a three-month period. Anthropic specifically noted that the queries targeted the model's most commercially valuable skills, including software engineering and agentic reasoning.
This campaign significantly dwarfs previous incidents reported by the company. In February, Anthropic alleged that several smaller Chinese startups—including DeepSeek, MiniMax, and Moonshot AI—had collectively generated over 16 million exchanges using about 24,000 fake accounts. The Alibaba operation alone exceeded the combined volume of those three earlier efforts.
National security and economic implications
Distillation is a process where users feed specific queries into a frontier AI model to collect responses, which are then used to train a smaller, less expensive model that mimics the original's performance. The U.S. government has already flagged this practice as a national security risk. Anthropic noted that the Alibaba campaign occurred after the White House issued a memo committing to share intelligence with labs regarding foreign distillation attempts.
The fallout from these allegations has been immediate for Alibaba:
- Alibaba's American depositary receipts fell more than 3 percent, dropping below $100 in trading.
- The company is currently facing pressure in Washington following its inclusion on the Pentagon's Chinese military companies blacklist.
- Anthropic warned that models built via distillation often lack necessary safety guardrails and allow rivals to replicate frontier AI at a fraction of the training cost.
Legislative responses and future outlook
In response to these findings, lawmakers are considering new measures to protect American intellectual property. Senators Bill Hagerty and Andy Kim plan to introduce an amendment to defense legislation that would blacklist or sanction Chinese firms found improperly accessing U.S. AI model outputs. Anthropic has urged the administration to clarify antitrust guidelines and implement penalties against firms engaging in systematic distillation.
While Alibaba has not commented on the specific allegations, the situation highlights a growing friction between American AI developers and international competitors seeking to bypass the massive costs of training foundational models from scratch.