Anthropic Accuses Alibaba of Stealing AI Models in Massive Distillation Attack

Anthropic has formally accused Chinese technology giant Alibaba of orchestrating one of the largest known adversarial distillation campaigns against its Claude AI model, revealing a months-long effort involving nearly 29 million exchanges from thousands of fraudulent accounts. The accusation comes at a delicate moment for US-China AI relations, as Washington tightens export controls while American labs face growing threats from rivals seeking to replicate frontier capabilities without the multi-billion-dollar R&D investment.

The campaign, detailed in a letter from Anthropic to US senators and White House officials, alleges that operators linked to Alibaba's Qwen AI lab systematically extracted Claude's most advanced capabilities in software engineering and agentic reasoning between April and June 2026. Anthropic estimates that roughly 25,000 fraudulent accounts were used to conduct 28.8 million exchanges with the model, calling it the largest attempt by a Chinese company to piggyback on US frontier AI. The practice, known as adversarial distillation, involves using a leading model's outputs to train rival systems at a fraction of the original development cost and violates most AI companies' terms of service.

Alibaba shares fell as much as 4.8% in Hong Kong following the news, which comes weeks after the US Defense Department added the company to a blacklist it is legally contesting. The accusation also arrives amid broader tensions between AI labs and the White House: Anthropic itself recently disabled access to its top two models, Fable 5 and Mythos 5, after US Commerce Department export restrictions, creating an ironic dynamic where the company is simultaneously calling for more government action against Chinese competitors while pushing back against its own regulatory constraints.

Adversarial distillation has become a flashpoint in the global AI race. OpenAI and Google have joined Anthropic in sharing information about similar attacks, and US senators Bill Hagerty and Andy Kim have introduced an amendment to a defence bill that would sanction firms found to be improperly accessing American AI output. White House Office of Science and Technology Policy Director Michael Kratsios published a memo in April specifically warning against the practice, which Anthropic claims Alibaba flagrantly disregarded.

Why it matters for Singapore: As a neutral hub that sits at the intersection of US and Chinese technology ecosystems, Singapore's AI governance framework will face increasing scrutiny over how companies operating here navigate the cross-border model-access wars. Singapore has positioned itself as a trusted environment for AI development and deployment, but the escalating conflict between US labs and Chinese rivals creates exposure for Singapore-based entities, research collaborations, and data centre operators that serve both markets. The city-state's voluntary governance model may need to evolve into firmer guardrails as adversarial distillation and related tactics become central national security concerns.