Anthropic has significantly reduced the extraction of internal reasoning tokens from its AI model Claude by 90 to 99 percent, responding to large-scale distillation attacks. According to The Decoder, this measure aims to protect the model’s intellectual property and prevent unauthorized replication of its reasoning capabilities.
The Decoder also reports that Anthropic has identified Chinese companies Zhipu and Moonshot as being linked to these industrial-scale attacks. In response, Anthropic banned tens of thousands of accounts involved in the misuse, aiming to contain the threat and safeguard its AI assets.
For Japanese markets, where AI-driven trading and analytics are increasingly important, this development underscores the broader challenges of protecting proprietary AI technology from industrial exploitation and highlights the growing tensions around AI security in the Asia-Pacific region.