Qihoo 360 unveils AI bug-hunting tools and claims parity with Anthropic's restricted Mythos model

Chinese cybersecurity company Qihoo 360 unveiled two AI vulnerability-hunting tools, Tulongfeng and Yitianzhen, at a Beijing conference on June 25, 2026, and its chief executive claimed they rival Anthropic’s export-restricted Mythos model. The pitch lands roughly two weeks after the US restricted Mythos and Fable 5, sharpening a contest over who supplies frontier security AI.

Tulongfeng is an AI agent that automatically builds sandbox environments, generates exploit code and runs real-world tests to confirm software flaws. Qihoo 360 said it has identified 3,432 vulnerabilities to date, 105 of them officially verified by Chinese authorities. The second tool, Yitianzhen, simulates attacks against an organization’s defenses and then suggests or carries out fixes.

Zhou Hongyi, Qihoo 360’s chief executive, said the tools can compete with Mythos on capability and framed vulnerability-detection AI as a national strategic asset, criticizing what he called ‘one-way transparency’ in global AI access. He also conceded that domestic base models still lag US systems by 20% to 30%, arguing his multi-agent approach closes the gap. Those performance claims are the company’s own and have not been independently benchmarked.

The timing is the story. Qihoo 360 unveiled the tools just after Washington blocked the public-facing Fable 5 and the more capable Mythos around June 12, citing fears the models could be jailbroken for offensive use. Chinese vendors are now positioning domestic alternatives directly against those restricted systems.

Whether Tulongfeng and Yitianzhen hold up under independent testing — or remain marketing against an export ban — will decide if Qihoo 360’s parity claim survives contact with outside researchers.