A Chinese cybersecurity company sanctioned by the US claims it has developed an artificial intelligence system capable of hunting software vulnerabilities on a scale comparable to Anthropic’s tightly controlled Mythos model, marking Beijing’s latest push to develop sovereign AI cyber capabilities.
Speaking at the ISC.AI 2026 cybersecurity conference in Beijing this week, Qihoo 360 chief executive Zhou Hongyi described Anthropic’s Mythos as a “cyber nuclear weapon” that has given the US a strategic advantage after export restrictions limited overseas access to the technology. He argued that China needed to develop its own equivalent rather than rely on access to American AI systems, Reuters reports.
Zhou said that rather than attempting to replicate Mythos directly, Qihoo had built a multi-agent platform known as Tulongfeng. The system combines multiple AI models with the company’s two decades of cybersecurity expertise, allowing different agents to collaborate on vulnerability discovery and analysis.
Zhou acknowledged that Chinese frontier AI models still lag behind leading US systems by around 20 to 30 percent in overall capability. Instead of trying to close that gap through model performance alone, he said Qihoo had focused on combining existing domestic AI models with specialist security tooling and workflows.
According to Zhou, the company has already used Tulongfeng to uncover more than 3,000 software vulnerabilities, 105 of which have been confirmed by Chinese authorities
Qihoo has not published technical evidence supporting those claims, and there has been no independent verification of the findings.
The announcement comes as Western intelligence and cybersecurity agencies warn that AI is rapidly changing the balance of cyber operations. Earlier this week, the Five Eyes intelligence alliance warned that adversaries could begin using AI to conduct sophisticated cyberattacks within months rather than years, raising concerns about the pace at which offensive capabilities are developing.
Qihoo 360’s claims are also likely to attract attention because of the company’s alleged relationship with the Chinese state. The firm has been on the US Commerce Department’s Entity List since 2020 over alleged links to China’s military. Earlier this month, the Pentagon identified Qihoo as a contributor to Beijing’s military-civil fusion strategy, describing it as part of China’s defence industrial base with ties to the country’s intelligence services. Qihoo has consistently denied the allegations.
Whether Tulongfeng can do everything Qihoo says it can is impossible to judge from the information released so far. What is clear is that China is investing heavily in home-grown cyber AI as access to the most capable American systems becomes increasingly difficult.
