Google’s AI security agents found 100+ critical software vulnerabilities in just two days

Mandiant disclosed AVDH, an internal multi-agent AI pipeline for automated vulnerability discovery that scanned tens of millions of lines of code and produced thousands of findings, including over 100 verified high-severity flaws and a dozen assigned CVEs; the report details the pipeline stages (threat modeling, entry-point discovery, context enrichment, hypothesis generation and validation), the human-in-the-loop validation process to reduce false positives, and guidance for defenders considering similar systems.