Cyber Risks · The Hacker News

Google's New AI Doesn't Just Find Vulnerabilities — It Rewrites Code to Patch Them

Summary: Google's DeepMind division announced CodeMender, an AI agent that detects, patches, and rewrites vulnerable code, built on Gemini Deep Think models with an LLM-based critique tool to check proposed changes for regressions. DeepMind said it has already upstreamed 72 security fixes to open source projects over six months, including some as large as 4.5 million lines of code, and Google also launched an AI Vulnerability Reward Program paying up to $30,000 for issues such as prompt injections, jailbreaks, and misalignment.

What you should know: Seventy-two fixes landed in open source projects in six months, and Google now plans to approach maintainers of critical projects directly with CodeMender-generated patches - so the code your supply chain depends on may soon change on an AI's recommendation, with a $30,000 bounty program signaling that AI-specific flaws are now a paid category of risk.

Read the original at The Hacker News ↗

Topics
AI Security
Related