Google DeepMind CodeMender: Autonomous Security Patching Agent Using Gemini Deep Think — 72 Patches Across 4.5M-Line Codebases

Google DeepMind·medium signal

Google DeepMind's CodeMender uses Gemini Deep Think models to autonomously find and fix security vulnerabilities. The agent combines static analysis, dynamic analysis, differential testing, fuzzing, and SMT solvers to trace root causes before generating patches. Over six months, CodeMender submitted 72 security patches to codebases up to 4.5 million lines. All patches are human-reviewed before upstream submission. Google plans to release CodeMender as a public developer tool.