Beijing Hits Back at Anthropic's Export Walls With Two AI Hacking Agents, One Handed Free to the World After the U.S. locked Anthropic's Claude Mythos behind export controls, Qihoo 360 founder Zhou Hongyi unveiled China's own vulnerability-hunting AI agents, while Z.ai released a rival model, GLM-5.2, free under an MIT license. Washington locked Anthropic's cybersecurity model Claude Mythos behind export controls and a carefully vetted coalition of partners. China's answer arrived on two fronts at once: a major industry conference where a prominent founder declared the country already had its own equivalent, and a separate lab that simply posted a comparable model online for anyone to download at no cost. Qihoo 360's pitch from the Beijing stage Speaking at ISC.AI 2026 in Beijing on June 24, Qihoo 360 founder Zhou Hongyi did not mince words: "China's cybersecurity industry must have its own Mythos." He used the platform to introduce Tulong Feng, an AI vulnerability agent that 360 is positioning as the country's homegrown version. He also rolled out Yitian Zhen, an automated defense platform, and launched a new domestic security alliance called "Panshi Zhidun," which translates to Shield of Bedrock. Why the founder calls Mythos a 'cyber nuclear weapon' Zhou's framing was deliberately sharp. In his view Mythos is the equivalent of "cyber nuclear weapons" in the AI era, an autonomous system capable of finding vulnerabilities, studying them, and assembling full attack chains with no human steering it. "U.S. organizations can use Mythos to scan your vulnerabilities, but you don't even have the right to look at Mythos," he said. He pointed to the fact that Chinese firms are shut out of Glasswing, Anthropic's vetted partner program whose members include Microsoft, Apple and other large technology companies. The track record Zhou claims for Tulong Feng By Zhou's account, Tulong Feng has so far uncovered a cumulative 3,432 vulnerabilities. Of those, 105 have been confirmed by Chinese regulatory bodies, and several were flagged as high-severity by the national vulnerability database. He argued that an agent-first strategy, one that coordinates several specialized models instead of betting on a single frontier system, makes up for whatever gap still separates China's base models from the best American ones. "America has Mythos," he told the room. "China also has its own 'Heaven-Sword Dragon-Saber.'" Z.ai makes its point by giving it away Beijing-based Z.ai, also known as Zhipu AI, took a very different route and proved its argument by shipping a product. The lab released GLM-5.2 soon after the U.S. government pulled Mythos 5 and Fable 5 offline for foreign nationals. GLM-5.2 is distributed under an MIT license, meaning there are no subscription gates, no geographic limits, and anyone is free to modify it. How GLM-5.2 performs on security tests The security benchmarks drew attention. In Semgrep's evaluation of insecure direct object reference detection, a test that measures whether a model can catch unauthorized object access flaws in code and is scored using the F1 metric that balances precision against recall, GLM-5.2 reached 39%, putting it ahead of Claude Code on the same test. A separate Graphistry evaluation found it level with Claude Opus 4.8 on a capture-the-flag challenge. The cost worked out to roughly $0.17 per finding, compared with more than $1 for Claude-based workflows. A direct reply to Musk's timeline Z.ai co-founder Tang Jie described Anthropic's withdrawal as "deeply regrettable," while the company's technical lead Qinkai Zheng was blunter: "We want the model accessible to everyone." When Elon Musk predicted that China would not match Fable-level capability until Q1 2027, Tang's response was short: "Won't take that long." What this means for you If you build software or run security for a living, this matters for both your toolkit and your budget. • For developers and security teams: An openly licensed, capable model like GLM-5.2 means powerful vulnerability-scanning tools are no longer locked behind export rules or vetted partner programs, and can be self-hosted and freely modified. • For cost-conscious teams: At roughly $0.17 per finding versus more than $1 for Claude-based workflows, the same security work could become far cheaper to run at scale. Questions & Answers 1. What did Qihoo 360 unveil? An AI vulnerability agent called Tulong Feng, an automated defense platform called Yitian Zhen, and a new domestic security coalition named Panshi Zhidun, or Shield of Bedrock. 2. Why does China want its own Mythos? Because Chinese companies are excluded from Anthropic's Glasswing partner program and cannot access Mythos, while U.S. organizations can use it to scan their vulnerabilities. 3. What is GLM-5.2 and what does it cost? It is Z.ai's model, released free under an MIT license with no geographic restrictions, and it costs roughly $0.17 per finding. 4. How does GLM-5.2 compare with Claude? It scored 39% on Semgrep's test, ahead of Claude Code, and matched Claude Opus 4.8 on a Graphistry capture-the-flag challenge. 5. How many vulnerabilities has Tulong Feng found? By Zhou's account, a cumulative 3,432, with 105 confirmed by Chinese regulatory bodies and several flagged as high-severity. 6. What did Elon Musk predict and how did Z.ai respond? Musk predicted China would not match Fable-level capability until Q1 2027, and Tang Jie replied that it would not take that long. https://trendkia.com/en/ai/anthropic-ki-pabndiyon-ke-javaba-men-china-ne-utare-do-eai-haikinga-ejenta-eka-to-puri-duniya-ke-lie-muphta-3657 TrendKia — Har trend, sabse pehle.