Anthropic and Chinese AI labs clash over chatbot distillation battle

This story crystallises how the AGI race is now as much about data and IP control as it is about raw model quality. Anthropic’s move to instrument Claude Code for signs of Chinese corporate usage, and the detailed allegations that labs like Alibaba’s Qwen, DeepSeek, Moonshot and MiniMax are distilling U.S. models at industrial scale, shows how frontier systems are being treated as strategic assets—not just software products. Distillation has been an accepted optimisation technique for years, but lab-on-lab distillation at this scale turns it into a geopolitical issue.

For the race to AGI, the key shift is that U.S. labs are explicitly asking Washington to treat aggressive distillation campaigns as a national security threat, not just a terms-of-service problem. That strengthens the hand of policymakers pushing for tighter export controls, KYC on model access, and perhaps even treaties on AI “intellectual re-use”, at the cost of fragmenting the global research commons. At the same time, the reporting makes clear that Chinese labs are quickly closing capability gaps using a mix of clever training tricks and lower prices, regardless of U.S. roadblocks.

The net result is a more polarised, adversarial landscape. U.S. labs may gain a temporary moat if they can truly starve off distillation, but they also risk overestimating how much IP control alone can slow Chinese innovation. Expect more monitoring, more restrictions, and more covert workarounds on both sides.