Google to Enhance Chrome Security with AI-Powered Updates Without Full Restart

Google announced plans to use neural networks to update Chrome without requiring a full restart, addressing the risk of zero-day exploits during the delay between patch release and user restart. The technology, called dynamic updating, replaces background child processes on the fly, leveraging Chrome's multi-layer architecture. Currently, Google continues development and will implement a two-week update cycle later this year, with Chrome 150 for macOS auto-restarting when updates are pending and windows are closed. Additionally, Google has been using large language models since 2023 to search for vulnerabilities in Chrome's codebase within a sandboxed environment with strict network restrictions. This effort led to the discovery of a bug that had persisted for 13 years. Google also collaborates with Google DeepMind and Project Zero on projects BigSleep and CodeMender, which helped identify over 20 vulnerabilities in May, including one critical. For fixing vulnerabilities, a multi-agent workflow is used, where language models generate patches, evaluate them, and create tests. Over the last two releases, Chrome 149 and 150, 1072 vulnerabilities were fixed, more than in the previous 23 releases combined.