Anthropic Accuses Chinese AI Firms of Illicitly Extracting Claude Capabilities

Anthropic, a prominent AI safety company, has formally accused three Chinese artificial intelligence firms of illicitly extracting capabilities from its advanced AI model, Claude. The accused companies are DeepSeek, Moonshot AI, and MiniMax. Anthropic states that these firms engaged in "distillation attacks," a sophisticated method of extracting proprietary AI model information.

The alleged attacks involved a massive scale of interaction, with over 16 million exchanges generated through the use of approximately 24,000 fraudulent accounts. These accounts were reportedly created to circumvent Anthropic's terms of service and bypass regional access restrictions that are in place to protect its intellectual property and ensure controlled deployment of its technology.

Anthropic has not yet detailed the specific technical methods used in the distillation attacks, nor has it specified the exact nature of the capabilities that were allegedly extracted. However, the company's statement indicates a serious breach of its security measures and terms of service. The use of a large number of fraudulent accounts suggests a coordinated effort to systematically probe and replicate Claude's functionalities.

This incident highlights ongoing concerns within the AI industry regarding the protection of proprietary models and the potential for unauthorized access and replication of advanced AI capabilities. The scale of the alleged operation, involving millions of exchanges and thousands of fake accounts, underscores the sophisticated methods that can be employed to target valuable AI assets.