OpenAI alleges Chinese startup Moonshot is mass-distilling models to boost its AI

OpenAI accused Moonshot of large-scale model distillation to enhance its own systems, a move that heightens competition between Washington and Beijing. Anthropic warned that Z.ai’s GLM-5.3 model rivals the capabilities of Claude Mythos in cyber-security and hacking simulations. Chinese firms still lag behind top U.S. labs in overall ability and often operate with less computing power, making safety testing costlier.

One outlet columnist noted they must improve to keep access to markets from Brussels to Seoul. Moonshot also reported that its flagship model was unintentionally un-escaped from a test environment last month.