China has its own mythology now, says Qihoo 360 founder. And one copy is free

The US has tied up Anthropic’s cybersecurity AI model behind Cloud Mythos Export controls And a A vetted coalition. China held a conference and announced that they already had something like it, and then a separate company uploaded roughly the equivalent onto the Internet for free.

Qihoo 360 founder Zhou Hongyi took the stage at ISC.AI 2026 in Beijing on June 24 and delivered a live message: “China’s cybersecurity industry should have its own mythology.” he unveil Tulong Feng – an AI 360 vulnerability agent called the Chinese version – along with Yitian Zhen, an automated defense platform, and a new local security alliance called “Panshi Zhidun” or Shield of Bedrock.

Cho’s framing was pointed. Mythos amounts to a “cyber nuclear weapon” for the age of artificial intelligence, an autonomous system that can find and analyze vulnerabilities and build attack chains without human guidance. “US organizations can use Mythos to scan your vulnerabilities, but you don’t even have the right to look at Mythos,” he said. Chinese companies are excluded from Glass pavilionAnthropic’s vetted partner program that includes Microsoft, Apple, and other big tech giants.

According to Zhou’s claims, Tulongfeng found a total of 3,432 vulnerabilities, of which 105 have been confirmed by Chinese regulatory agencies and several of which have been flagged as high-risk by the national vulnerability database. Zhou claimed that the agent-first route – coordinating specialized models rather than betting on a single parametric system – makes up for any basic model gap that still exists. “America has myths,” he told the audience. “And China also has its own Sky Sword Dragon Sword.”

Based in Beijing Z.ai He also made a different argument by launching one. The laboratory, also known as Zhipu AI, shot down GLM-5.2 shortly after the US government Mythos 5 and Fable 5 pulled Currently for foreign citizens. GLM-5.2 runs under the MIT license, there are no subscription portals, no geographic restrictions, and anyone can freely modify it.

The cybersecurity numbers were notable. Semgrep’s evaluation of insecure direct object reference detection—a test that measures whether a model can detect unauthorized object access flaws in code, scored via an F1 metric that balances precision and recall—placed the GLM-5.2 at 39%, ahead of Claude Code on the same test.

A separate graphical evaluation found it matched Cloud Opus 4.8 in the Capture the Flag Challenge. Cost per result: About $0.17, versus more than $1 for Cloud-based workflow.

Tang Jie, co-founder of Z.ai, described Anthropic’s withdrawal as “Very unfortunateThe company’s technical lead Qinkai Zheng was more blunt: “We want the model to be affordable for everyone.” When Elon Musk predicted that China wouldn’t match Fable-level capability until the first quarter of 2027, Tang responded: “It won’t take that long.”