A Chinese AI security system named Tu Long Feng, developed by 360 Security, has achieved a record-breaking 99.4% success rate on the CyberGym leaderboard, a benchmark for AI security capabilities. This achievement surpasses other leading AI models such as OpenAI's GPT 5.5 Cyber, Anthropic's Claude Mythos Preview, and Google's Gemini 3.8 Flash Cyber. The system's success is attributed to its training on extensive real-world vulnerability data and expert human experience, enabling it to effectively identify, exploit, and reproduce complex security flaws. AI
IMPACT Sets a new benchmark for AI capabilities in specialized security tasks, potentially influencing future AI development in cybersecurity.
RANK_REASON AI system from a non-frontier lab achieves SOTA on a specialized benchmark, surpassing frontier models. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
- 360 Security
- Anthropic
- Claude 3.7 Sonnet
- Claude Mythos Preview
- Gemini 3.8 Flash Cyber
- GPT 5.5 Cyber
- Master of Pwn
- National Vulnerability Database
- OpenAI
- OpenHands
- Pwn2Own
- Tu Long Feng
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
