China Has Its Personal Mythos Now, Says Qihoo 360 Founder. And One Model Is Free – Decrypt

The U.S. restricted Anthropic’s cybersecurity AI mannequin Claude Mythos behind export controls and a vetted coalition. China held a convention and introduced it already has one thing prefer it—then a separate firm uploaded a near-equivalent to the web free of charge.

Qihoo 360 founder Zhou Hongyi took the stage at ISC.AI 2026 in Beijing on June 24 and delivered a direct message: “China’s cybersecurity trade should have its personal Mythos.” He unveiled Tulong Feng—an AI vulnerability agent 360 is looking China’s model—alongside Yitian Zhen, an automatic protection platform, and a brand new home safety coalition named “Panshi Zhidun,” or Defend of Bedrock.

Zhou’s framing was pointed. Mythos quantities to “cyber nuclear weapons” within the AI age—an autonomous system that may discover vulnerabilities, analyze them, and construct assault chains with out human path. “U.S. organizations can use Mythos to scan your vulnerabilities, however you do not even have the precise to have a look at Mythos,” he stated. Chinese language firms are excluded from Glasswing, Anthropic’s vetted associate program that features Microsoft, Apple, and different large tech giants.

In line with Zhou’s claims, Tulong Feng has discovered a cumulative 3,432 vulnerabilities, with 105 confirmed by Chinese language regulatory our bodies and a number of other flagged as high-severity by the nationwide vulnerability database. Zhou claimed the agent-first route—coordinating specialised fashions quite than betting on a single frontier system—offsets no matter base-model hole nonetheless exists. “America has Mythos,” he advised the viewers. “China additionally has its personal ‘Heaven-Sword Dragon-Saber.'”

Beijing-based Z.ai additionally made a special argument by releasing one. The lab, also referred to as Zhipu AI, dropped GLM-5.2 shortly after the U.S. authorities pulled Mythos 5 and Fable 5 offline for overseas nationals. GLM-5.2 runs below an MIT license—no subscription gates, no geographic restrictions, freely modifiable by anybody.

The cybersecurity numbers had been notable. Semgrep’s analysis of insecure direct object reference detection—a take a look at measuring whether or not a mannequin can spot unauthorized object entry flaws in code, scored by way of the F1 metric that balances precision in opposition to recall—put GLM-5.2 at 39%, forward of Claude Code in the identical take a look at.

A separate Graphistry analysis discovered it matched Claude Opus 4.8 on a capture-the-flag problem. Price per discovering: round $0.17, versus over $1 for Claude-based workflows.

Z.ai co-founder Tang Jie known as Anthropic’s withdrawal “deeply regrettable.” The corporate’s technical lead Qinkai Zheng was extra blunt: “We wish the mannequin accessible to everybody.” When Elon Musk predicted China would not match Fable-level functionality till Q1 2027, Tang replied: “Will not take that lengthy.”