Anthropic Accuses Chinese Labs of Distilling Claude as US Debates AI Chip Exports

Anthropic accuses Chinese labs of using 24,000 fake accounts to distill Claude's capabilities, as the United States debates AI chip exports.

Anthropic Accuses Chinese Labs of Distilling Claude as US Debates AI Chip Exports
DeepSeek logo on a mobile screen with the Chinese flag and binary code

In February 2026, Anthropic accused three Chinese AI labs, DeepSeek, Moonshot AI and MiniMax, of creating more than 24,000 fake accounts to interact with its Claude model, generating more than 16 million exchanges. This process, known as "distillation", was used to extract Claude's distinctive capabilities, such as agentic reasoning, tool use and programming, in order to improve the labs' own models. These actions raise concerns about the protection of intellectual property and national security, as they could allow state or non-state actors to develop advanced technologies without the safeguards implemented by Anthropic. (techcrunch.com)

Distillation is a common technique in training AI models, in which a smaller model learns from a larger one. Although it is a legitimate practice, the unauthorized use of other companies' models can violate terms of service and intellectual property laws. Anthropic found that DeepSeek focused its efforts on improving foundational reasoning and alignment, Moonshot AI targeted agentic reasoning and tool use, while MiniMax directed its work toward programming and orchestration. These attacks raise questions about the need for stricter regulation to protect innovation in the AI field. (tomshardware.com)

Implications for US Policy on AI Chip Exports

These accusations come at a critical moment, as US authorities are considering easing export controls on advanced AI chips to China. The unauthorized distillation of models such as Claude by Chinese labs highlights the risks associated with unregulated access to advanced technologies. Anthropic has stressed that access to advanced chips is essential to carry out large-scale distillation attacks, strengthening the case for keeping strict chip export controls. (techcrunch.com)

Reactions and Future Outlook

Anthropic's accusations have drawn mixed reactions. Dmitri Alperovitch, chairman of the Silverado Policy Accelerator think tank, said it was not surprising to see such attacks, since the rapid progress of Chinese AI models has been partly fueled by the theft of US models through distillation. This should provide even stronger reasons to refuse the sale of AI chips to these companies, as it would further benefit them. (techcrunch.com)

In response, Anthropic is investing in defenses to make distillation attacks harder and easier to identify, and is calling for a coordinated response across the AI industry, cloud providers and policymakers. The situation underlines the need to balance promoting innovation with protecting intellectual property and national security. (tomshardware.com)

Takeaway

  • Protecting intellectual property: Companies must implement advanced security measures to protect their AI models from unauthorized access and illicit distillation.
  • Export regulation: Governments need to set clear policies on the export of advanced technologies to prevent unauthorized access by foreign actors.
  • International collaboration: The global AI community should work together to develop ethical and legal standards that govern the use and sharing of AI models.
  • Awareness of risks: Companies must be aware of the risks associated with using third-party AI models and carefully assess the legal and security implications.
  • Responsible innovation: It is essential to promote AI innovation responsibly, ensuring that technologies are used for beneficial purposes and not for harmful activities.

For further insights on this topic, you can watch the following video:

Your browser does not support the video tag.