CodeMender, Google Cloud, adds controls for isolating AI vulnerability scans

Businesses can now direct code scans with greater accuracy and validate discovered weaknesses by utilizing new controls integrated into CodeMender, Google Cloud's artificial intelligence utility designed for locating and resolving software flaws. These enhancements permit organizations to conduct attack simulations within a controlled environment, either an isolated virtual machine or a sandbox operating system level. Furthermore, to ensure testing remains within established boundaries, the system incorporates human oversight for actions such as executing commands and making local writes. The functionality allows for comprehensive scanning across an entire code repository, or the ability to focus on a particular section. This development stems from a post originally published on Help Net Security, detailing how Google is imposing refined constraints on AI agents tasked with threat hunting and vulnerability remediation.