Project Glasswing: what Mythos showed us
Summary
Cloudflare analyzes Mythos Preview within Project Glasswing, exploring how AI-powered vulnerability research harnesses LLMs to chain exploits, generate PoCs, and improve triage. The post covers model refusals, the need for a harness to scale vulnerability discovery, and the staged approach to harnessing AI for security testing, with implications for how security teams patch and defend systems.