The 'LLMs as Offensive Agents' Benchmark

A foundational study proving that models can autonomously conduct reconnaissance and exploit software vulnerabilities without human intervention.

Sources

Open the full topic