autonomous patching

Google DeepMind introduces CodeMender, an AI agent autonomously identifying and fixing security vulnerabilities in code. Over six months, it contributed 72 security patches to open-source projects. CodeMender proactively rewrites existing code to eliminate security flaws, using Gemini models for analysis and a validation process. This approach hardens software against future threats and addresses complex issues typically missed, significantly improving open-source software security. A cautious deployment with human oversight ensures reliability and quality. DeepMind aims to release CodeMender as a public tool, sharing research to enhance software security.