Anthropic Accuses Chinese AI Labs of Secretly Training on Claude Data
Anthropic says China-based AI firms including Alibaba and Moonshot AI covertly used millions of Claude interactions to train competing models.
Anthropic on Monday accused several China-based artificial intelligence laboratories of secretly harvesting millions of conversations with its Claude AI models to train their own competing systems, marking a significant escalation in tensions over intellectual property in the global AI race. The San Francisco-based company said it detected unauthorized activity it attributed to firms including Alibaba and Moonshot AI, two of China's most prominent players in the generative AI space.
The alleged scheme involved using Claude's outputs at scale — without Anthropic's permission — to effectively distill the model's capabilities into rival systems, a technique sometimes called "model distillation" or "data harvesting." Such practices violate Anthropic's terms of service and raise broader questions about how frontier AI models can be exploited to accelerate competitors' development pipelines at minimal cost.
Read more Trust Stamp Broadens ID Dataweb Partnership Across Enterprise Base →
The disclosure puts a spotlight on a vulnerability facing virtually every major AI provider: sophisticated actors can, in theory, systematically query public-facing models and use the responses as training data to approximate those models' performance. For Anthropic, which has positioned Claude as a safety-focused alternative to OpenAI's ChatGPT and Google's Gemini, the breach represents both a reputational and a competitive challenge at a critical moment in the industry's growth.
The incident also arrives against a backdrop of intensifying US-China rivalry over AI supremacy, with Washington imposing export controls on advanced chips and Beijing pushing domestic labs to close the capability gap with Western counterparts. Whether Anthropic pursues legal action or additional technical countermeasures against the identified labs remains to be seen, but the revelation is likely to prompt other frontier AI companies to reassess how they monitor and restrict access to their own models.
Continue reading at US Top News and Analysis.
