Google's AI Finds 100+ Vulnerabilities in 48 Hours

Google's Threat Intelligence Group has released the blueprint for an agentic AI system. This system, called AVDH, helped Mandiant find over 100 critical software vulnerabilities in just 48 hours. What's interesting is that this internal AI pipeline automates deep security reviews of source code. It helps defenders find weaknesses before attackers can exploit them. Google is sharing these details to help security teams develop similar approaches to protect software. This move comes as threat actors increasingly use AI, raising the risks of source code theft. AVDH has been used by Mandiant for the past ten months in various security operations. During one incident involving stolen code, it found more than 100 critical vulnerabilities in two days. The technology has also analyzed millions of lines of code, uncovering dozens of flaws in widely used web extensions and open-source projects, leading to assigned CVEs. The system uses a multi-agent approach, with specialized AI agents performing different stages of vulnerability discovery and validation. This allows for a thorough and automated review process. This innovation could significantly strengthen our defenses against sophisticated cyber threats.