Google’s Wiz Unveils Atlas AI System for Vulnerability Research

Wiz, the cloud-security company acquired by Google for $32 billion in March 2026, built Atlas to turn costly, point-in-time vulnerability research into scalable AI code scanning. The system assigns specialized agents to map attack surfaces, hunt for flaws in parallel, challenge one another’s findings and reproduce exploits in dedicated environments. The approach matters because it aims to reduce false positives while making deep security analysis economical enough to run continuously across large codebases.

On July 27, 2026, Wiz said Atlas ranked first on the CyberGym benchmark with a 90.9% success rate and had uncovered more than 200 previously unknown flaws in projects including Kubernetes, the Linux kernel and containerd. An early version also helped researchers identify CVE-2026-3854, a critical remote-code-execution flaw in GitHub.com and GitHub Enterprise Server, reported on March 4. GitHub mitigated the issue on GitHub.com within six hours and awarded a $100,000 bounty, the largest in the program’s history.