Chinese AI startup's GLM-5.3 outperforms Anthropic model in cybersecurity test

Zhipu's GLM-5.3 release signals continued momentum in China's open-weight AI development, even as geopolitical tensions around AI innovation intensify. The model's claimed parity with Anthropic's Mythos 5 in cybersecurity testing and progress on coding follows Zhipu's June release, demonstrating a pace of iteration that the startup credits to the competitive landscape. The timing coincides with a hardening of US policy: the White House is reportedly crafting plans to force countries into explicit alignment with the US in AI competition, effectively blocking participation in Chinese frameworks. However, the Trump administration's simultaneous easing of restrictions on certain US chip exports to China suggests policy inconsistency—a point a Council on Foreign Relations expert explicitly flagged as "incoherent." This contradiction may reflect internal debate within the administration about how aggressively to constrain China's access to technology versus maintaining economic relationships.