Anthropic Alleges Largest-Ever Claude Distillation Attack by Alibaba
Anthropic has accused Chinese company Alibaba of conducting a large-scale Claude distillation attack, allegedly using Anthropic's AI models to train competing systems. The company claims this is the largest such attack ever detected.
Background
- Anthropic, the AI company behind the Claude family of large language models (LLMs), has publicly accused Alibaba Group of orchestrating a large-scale "model distillation" attack.
- Model distillation is a technique where a smaller, cheaper model is trained to mimic a larger, more capable model by feeding it outputs from the target model. While legal in some contexts (e.g., open-source models), using it against a proprietary, commercial model to extract capabilities without permission is generally considered a violation of terms of service and, in some jurisdictions, intellectual property or anti-circumvention law.
- Anthropic claims this was the largest known such attack against Claude, involving a high volume of API requests designed specifically to extract Claude's behavior and knowledge into a competing Alibaba model.
- This escalates an ongoing tension between Western AI labs (Anthropic, OpenAI, Google, Meta) and Chinese tech giants over data scraping, IP theft, and the security of frontier AI models. Alibaba, as one of China's largest cloud and AI players, develops its own competing LLM series, Qwen.