The landscape of advanced Artificial Intelligence is rapidly evolving, particularly in the critical domain of cybersecurity. This week saw significant developments, with Chinese tech firm Qihoo 360 unveiling its indigenous AI vulnerability-hunting system, “Tulong Feng.” Simultaneously, Z.ai (Zhipu AI) released its GLM-5.2 model as open-weight code, offering comparable capabilities to proprietary systems but with complete accessibility. These moves occur against a backdrop of international discussions surrounding export controls on advanced AI models, specifically concerning Anthropic’s cybersecurity AI, Mythos. The unfolding events highlight a competitive push in developing autonomous AI agents capable of identifying and analyzing software vulnerabilities, a key area for both offensive and defensive cybersecurity strategies. As global powers and tech companies refine these sophisticated tools, the implications for blockchain security, Layer 2 solutions, and the broader Web3 ecosystem are profound, potentially accelerating the development of more robust decentralized applications and infrastructure. Key Takeaways
- Qihoo 360 introduced “Tulong Feng,” an AI vulnerability agent, positioning it as China’s answer to sophisticated foreign AI models in cybersecurity.
- Z.ai released GLM-5.2 under an MIT license, making its advanced vulnerability detection capabilities freely available to the public.
- GLM-5.2 demonstrated competitive performance against proprietary models like Claude Code on specific vulnerability detection benchmarks.
- The accessibility of Z.ai’s model challenges the exclusive control of advanced AI tools, potentially democratizing cybersecurity research.
- These developments underscore the growing importance of AI in securing digital infrastructure, including the rapidly expanding Web3 and blockchain sectors.
Qihoo 360’s founder, Zhou Hongyi, presented Tulong Feng at the ISC.AI 2026 conference in Beijing. He framed the AI as essential for China’s cybersecurity industry, emphasizing the need for domestic solutions in the face of international restrictions on advanced AI models like Anthropic’s Mythos. Zhou described these AI systems as “cyber nuclear weapons,” capable of autonomous vulnerability discovery and exploitation chain development. The restricted access to such tools for Chinese entities spurred the creation and announcement of Tulong Feng, alongside an automated defense platform, Yitian Zhen, and a domestic security coalition, “Shield of Bedrock.” According to Qihoo 360’s claims, Tulong Feng has identified over 3,400 vulnerabilities, with a significant portion confirmed by national regulatory bodies and flagged as high-severity. Zhou suggested that an “agent-first” approach, orchestrating specialized AI models, can compensate for any perceived gaps compared to monolithic frontier models. In parallel, Z.ai’s release of GLM-5.2 as open-weight code offers a different paradigm. The model, available under the permissive MIT license, bypasses subscription gates and geographic restrictions, allowing any developer to download, modify, and utilize it. This move directly contrasts with the controlled access model of proprietary AI systems.
Long-Term Technological Impact on Cybersecurity and Web3
The dual advancements from Qihoo 360 and Z.ai, particularly the open-sourcing of GLM-5.2, carry significant long-term implications for the blockchain and Web3 industries. The availability of powerful, AI-driven vulnerability detection tools outside of corporate or governmental control can dramatically democratize security auditing. For blockchain developers building Layer 2 solutions, decentralized applications (dApps), and smart contracts, having access to such AI can mean more proactive and comprehensive security assessments. This could lead to a significant reduction in exploits and hacks that have plagued the space, fostering greater trust and adoption. Furthermore, the competitive pressure generated by these developments may accelerate innovation in AI-native security solutions specifically tailored for blockchain environments. We might see the emergence of specialized AI agents designed to detect novel vulnerabilities in smart contract code, identify consensus mechanism weaknesses, or even predict and mitigate network-level attacks. The open-source nature of GLM-5.2 also encourages community-driven improvements and adaptations, potentially leading to AI tools that are more robust, adaptable, and effective at securing the decentralized future. The strategic release of these technologies signals a growing recognition of AI’s indispensable role in maintaining the integrity and security of evolving digital ecosystems.
Original article : decrypt.co