360's self-developed AI security attack-defense system Tulongfeng tops CyberGym with a 99.4% vulnerability reproduction success rate

360's self-developed AI security attack-defense system Túlóngfēng topped CyberGym with a 99.4% vulnerability reproduction success rate and set a new record; the same leaderboard includes models and agents from OpenAI, Anthropic, and Google. The system underwent three rounds of independent evaluation across 1507 tasks, with an average success rate of 99.38%, adopting a "security-domain large model + specialized agent swarm" approach. It has cumulatively discovered over 15000 vulnerabilities, and more than 500 enterprises have onboarded.