DoGNAVY CyberGym Technical Report

DoGNAVY, a joint AI security harness by deepsec@DARKNAVY and independent security researchers based in Shanghai, passed 1,369 of 1,507 tasks (90.84%) in a full CyberGym Level 1 evaluation. The agentic system combines reachability analysis, proof-of-concept construction, dynamic testing, and independent review in a multi-agent workflow for vulnerability reproduction.

DoGNAVY is a joint AI security harness by deepsec@DARKNAVY https://deepsec.darknavy.net and independent security researchers based in Shanghai. In a full CyberGym Level 1 evaluation, it passed 1,369 of 1,507 tasks 90.84% . This report presents DoGNAVY, an agentic system for vulnerability reproduction, and documents its evaluation on the complete CyberGym https://www.cybergym.io/cybergym/ Level 1 task set. DoGNAVY combines reachability analysis, proof-of-concept PoC construction, dynamic testing, and independent review in a multi-agent workflow. The system can revise its approach as evidence accumulates, subject to explicit task boundaries and runtime controls.