A Chinese cybersecurity firm, 360 Digital Security Group (Qihoo 360), claims its AI-driven Multi-Agent Collaborative Vulnerability Discovery System can find vulnerabilities at a scale approaching Anthropic’s Claude Mythos. ETH Zurich researcher Eugenio Benincasa’s analysis finds 360’s system produced many contest findings and notable CVE claims but stops short of Mythos-level reasoning and highlights China’s reporting rules that channel research to state agencies. #ClaudeMythos #Qihoo360
Keypoints
- 360 Digital Security Group says its multi-agent system helped identify nearly 1,000 vulnerabilities, including over 50 high-severity flaws.
- The system reportedly contributed to about half of 360’s findings at the Tianfu Cup, helping the company secure first place.
- 360 claims rapid discovery of CVE-2026-32190 after years undetected, while attribution for CVE-2026-24293 is disputed.
- Benincasa assesses 360’s capabilities as significant but not yet matching the autonomous reasoning Anthropic attributes to Claude Mythos.
- Chinese rules requiring vulnerability reporting to government agencies may channel elite research into state intelligence pipelines, creating a strategic advantage.