⚡️AI UPDATE: OpenAI ChatGPT 5 and DeepSeek V4 Flash records HIGH success rates in cyber security tasks. CyberGym's test results show AI is rapidly moving from simply writing code to autonomously finding and reproducing real security vulnerabilities. The benchmark includes: - 1,507 real vulnerabilities - 188 software projects - Complex exploit and patching tasks The original top score was around 20%. Newer specialized agents reportedly exceed 90% under updated testing conditions. AI can now: - Reproduce known vulnerabilities - Discover zero-days - Identify incomplete patches Interestingly, the SAME technology can strengthen cyber defenses or scale attacks.(Coin Bureau)