A turning point on AI agents in Cybersecurity, shown in two recent research papers from UC Berkeley and Stanford:

  CyberGym:  AI agents discovered 15 zero-days in major open-source software

  BountyBench:  AI agents solved real-world bug bounty tasks worth tens of thousands of dollars

This represents a pivotal shift in cybersecurity — AI agents can now autonomously do what only elite human hackers could before.

Check out their work:

CyberGym: https://www.cybergym.io/

BountyBench: https://bountybench.github.io/