LESSONS LEARNED BUILDING AN AI HACKER

Summary from the TLDR InfoSec newsletter

This post dives into how Theori built effective LLM agents for DARPA's AI Cyber Challenge. The agents worked by breaking problems into sub-tasks, carefully curating narrow, guarded tools and enforcing structured outputs while specialized sub-agents handled recurring jobs like source questions and PoV debugging.