CodeMender is an AI agent introduced by Google DeepMind in October 2025 that automatically detects security vulnerabilities in software code and generates patches to fix them. Built on Gemini models, it analyzes code and uses tools such as static analysis, fuzzing, and tests to trace the root cause of a flaw.
Patches produced by the agent are reviewed by human researchers before being contributed to open-source projects. Beyond fixing known flaws, the project also aims to proactively rewrite code into more secure forms.