Detailed Analysis
Anthropic, the American AI safety company behind the Claude family of large language models, has publicly accused Chinese technology giant Alibaba of illicitly extracting AI capabilities from its systems. The accusation places Anthropic among a growing number of Western AI developers alleging that Chinese competitors have engaged in unauthorized methods to obtain proprietary model capabilities, a practice that typically involves using outputs from a target model to train or enhance a rival system — a technique broadly known as model distillation or knowledge extraction. Alibaba, which operates its own suite of AI models under the Qwen brand, has emerged as one of China's most aggressive competitors in the global foundation model market.
The allegation reflects a deepening tension at the intersection of AI development and geopolitical competition. The practice of extracting capabilities from frontier models without authorization raises significant legal and ethical questions around intellectual property, terms of service violations, and competitive fairness. Anthropic, which has positioned itself as a safety-focused lab and has received substantial investment from companies including Google and Amazon, has a strong commercial and reputational interest in protecting the integrity of its Claude models from exploitation by competitors who could benefit from Anthropic's extensive training investments without bearing the associated costs.
This development connects to a broader pattern that has drawn regulatory and industry attention across the AI sector. OpenAI has previously alleged that DeepSeek, another Chinese AI developer, used distillation techniques to extract capabilities from ChatGPT, and multiple major AI labs have strengthened their terms of service and detection mechanisms in response to suspected model scraping and output harvesting. The ease with which capable models can theoretically be queried at scale to generate synthetic training data makes enforcement challenging, even as it becomes a central concern for frontier AI developers seeking to maintain competitive advantages.
The accusation also highlights the structural vulnerabilities inherent in deploying powerful AI models through public APIs and consumer products. While such deployment is commercially necessary, it creates vectors through which competitors can systematically probe model behavior and use that data to bootstrap or accelerate their own development. The Anthropic-Alibaba dispute may ultimately push the industry toward more robust technical countermeasures, stronger legal frameworks around AI output ownership, and potentially renewed calls for international norms governing the use of competitor model outputs — questions that governments and standards bodies have only begun to formally address.
Read original article →