Chinese AI systems are matching the performance of Anthropic's top model Mythos in some cybersecurity scenarios, putting pressure on U.S. AI policy, the Wall Street Journal reported on June 27 local time.

Security researchers said GLM-5.2, unveiled in June by China's Z.ai, shows performance on par with the latest U.S. models in detecting software bugs. GLM-5.2 ranked among the top 10 most-used AI models in OpenRouter data, and in the Semgrep benchmark test it sometimes outperformed Anthropic's Claude Opus 4.8.

With additional instructions, both Opus 4.8 and GLM-5.2 show bug-detection capabilities comparable to Mythos, the researchers said.

GLM-5.2 is an open-weight model, allowing anyone to freely download and modify it.

360 Security Technology also unveiled a bug-detection tool, Tulongfeng, on June 24 that it said matches Mythos. Zhou Hongyi (저우훙이), chief executive of 360 Security, said at a Beijing conference, "Such weapons that could change the landscape of the cyber battlefield should not belong only to the United States."

China's AI advances are unfolding as the U.S. government applies unprecedented brakes on the release of AI models.

OpenAI limited access to its latest model, GPT-5.6, on June 26 due to security concerns. Anthropic's Fable 5 and Mythos 5 were blocked for more than two weeks after the Trump administration ordered a ban on foreign users, before some access to Mythos 5 was restored on June 26.

Saif Khan (사이프 칸), a researcher at the Institute for Progressive Studies who handled export controls under the Biden administration, criticised the approach, saying, "Blocking Fable while exporting chips China needs to develop its own models is giving China a gift."

Jacob Helberg (제이콥 헬버그), an undersecretary of state, said, "We are closely tracking Chinese open-source models."

Niels Provos (닐스 프로보스), a researcher who led security teams at Google and Stripe, said, "It is luring companies around the world to use cheaper but highly capable Chinese open-weight models while weakening the U.S. AI industry," adding, "I cannot understand it."