Anthropic Accuses Alibaba of Large-Scale Distillation Attack on Claude

Core View

Anthropic alleges Alibaba conducted a ‘largest known’ distillation attack on its Claude AI model by creating nearly 25,000 fake accounts.
The attack involved over 29 million interactions with Claude to extract capabilities in areas like agent reasoning, software engineering, and long-term task management.
Claude does not allow access to Chinese entities, making the attack a violation of its access policies and security protocols.
[AI Synthesis] This incident highlights growing concerns about AI model misuse, particularly through unauthorized data extraction and model distillation by foreign entities.

The accusation is based on a letter obtained by The Wall Street Journal, addressed to U.S. Senators Tim Scott and Elizabeth Warren, dated June 10, 2026.
Anthropic claims Alibaba’s actions represent a significant threat to U.S. AI investment and R&D, converting billions in American investment into ‘subsidies’ for competitors.
This is not the first time Anthropic has raised concerns — in February, it reported industrial-scale distillation activities by three Chinese AI labs using Claude.

[AI Synthesis] The attack raises concerns about the security of frontier AI models and the potential for geopolitical manipulation of AI development trajectories.
Anthropic calls for government-industry coordination to protect U.S. AI leadership, including tighter chip controls and legislation to punish entities involved in distillation attacks.

Alibaba allegedly used over 25,000 fake accounts to extract capabilities from Claude through over 29 million interactions.
The attack targets high-value AI functions such as agent reasoning and long-term planning, indicating a strategic effort to replicate advanced AI behaviors.
This incident underscores the need for stronger AI model access controls, international cooperation, and legal frameworks to prevent unauthorized model distillation.

Topics: Tech
Tags: ai-security model-distillation claude alibaba