At the ISC.AI 2026 conference in Beijing, 360 Security Technology founder Zhou Hongyi unveiled a new AI-powered tool for automated software vulnerability discovery — Tulong Feng. This development is positioned as a direct Chinese response to Anthropic's Mythos 5, highlighting the growing competition in the fields of cybersecurity and AI.

Tulong Feng is part of the broader Yitian Tulong suite, which also includes the Yitian Zhen automated protection system. In addition, 360 has initiated the creation of the "Basalt Shield" coalition, aimed at protecting companies operating in critical infrastructure and the import-substituting IT solutions sector.

According to Zhou, Tulong Feng has already demonstrated impressive results, discovering 3,432 software vulnerabilities. Of these, 105 were officially confirmed by Chinese regulatory authorities, and several were added to the national database with a "highly critical" label. It is worth noting that independent verification of this data has not yet been conducted, leaving room for discussion.

In his speech, Zhou emphasized that China should not blindly copy the American strategy based on creating "the most powerful models, the most efficient computing, and the most advanced chips." Instead, 360 is betting on AI agents that integrate language models, the expertise of security specialists, and extensive vulnerability knowledge bases. This approach, in my view, is more pragmatic and adaptive given the current geopolitical and technological constraints.

This event occurs against the backdrop of tightening regulations in the United States. Previously, the U.S. government demanded that Anthropic suspend access to the Fable 5 and Mythos 5 models for foreign nationals, leading to a complete shutdown of these models for all clients. The reason, according to some reports, was White House concerns about potential access to the Mythos model by a Chinese group. In response to these restrictions, OpenAI also stepped up its efforts, launching the full version of the specialized GPT-5.5-Cyber model for vulnerability discovery and patching, and then opening limited access to the GPT-5.6 Sol, Terra, and Luna models for a narrow circle of trusted partners.

It is worth noting that earlier, Chinese startup Zhipu AI released its flagship open-source model GLM-5.2, focused on long agent tasks and programming. This points to a systemic trend: China is actively building its own AI capabilities, aiming to reduce dependence on Western technologies and create an autonomous ecosystem.

Expert opinion: The launch of Tulong Feng is not just a technological novelty, but a clear signal of the formation of a new landscape in cybersecurity, where AI agents will become a key tool. China is betting on the practical integration of knowledge and an agent-based approach, which could give it an advantage in flexibility compared to the cumbersome but powerful American models. In the coming years, we will likely witness a true AI arms race in the field of vulnerability discovery.