Autonomous fuzzing process under LLM supervision

Fuzzlab is a proof-of-concept, LLM-supervised fuzzing pipeline that automates preparation, test-harness generation, crash triage, ML-driven prioritization, and reporting; the report details its architecture, telemetry layers, self-healing mechanisms, and ML/LLM integrations and cites two disclosed findings — CVE-2026-35251 (VirtualBox IOMMU emulation flaw enabling potential VM escape) and CVE-2026-42268 (ModSecurity integer overflow causing web server DoS) — as examples of real vulnerabilities the system helped find.