China Considers Restricting Access to Top AI Models Amid Concerns Over 'Distillation'
The move comes as Chinese AI firms are accused of using U.S. models to improve their own, raising national security and commercial worries.
China is reportedly considering new regulations that would restrict access to its most advanced artificial intelligence models, a move that comes as U.S. companies accuse Chinese firms of illicitly using their proprietary AI to enhance their own systems. The proposed restrictions aim to address concerns over a practice known as "distillation," where the outputs of a more powerful AI are used to train a less sophisticated one.
Chinese AI models are said to lag behind leading American counterparts by approximately seven months on key performance benchmarks. To bridge this gap, companies like Anthropic allege that Chinese laboratories are engaging in distillation, a practice that violates their terms of service. Anthropic reported in February that firms including DeepSeek, Moonshot, and MiniMax generated an estimated 16 million interactions with its Claude model using roughly 24,000 fraudulent accounts.
Further fueling these tensions, Anthropic reportedly sent a letter to U.S. officials in June accusing Alibaba of "brazenly" attempting to replicate Claude's capabilities. The company stated it had developed code specifically designed to detect and flag such unauthorized usage.
From Anthropic's perspective, distillation poses significant national security risks. The company warns that foreign entities could leverage this method to transfer AI capabilities into military and surveillance systems, potentially enabling authoritarian governments to "deploy frontier AI for offensive cyber operations, disinformation campaigns, and mass surveillance." Additionally, there is a substantial commercial interest at stake. If Chinese firms can siphon U.S. AI capabilities and release them as open-source, it could undermine the business models of the U.S. companies developing these advanced technologies.