Google Says AI Agent Helped Find 13-Year-Old Chrome Sandbox Escape
Summary
Google says its Gemini-based agent helped uncover a 13-year-old Chrome sandbox-escape flaw as the browser’s security fixes surged. The company also outlined faster release plans and broader memory-safety efforts.
Key points
- Google says an agent harness using Gemini helped identify security flaws across Chrome’s codebase and contributed to a sharp rise in fixes.
- Google patched 1,072 security defects in Chrome 149 and 150, exceeding the combined total for the previous 23 milestones, according to the company.
- The AI-assisted effort found CVE-2026-3545, a 13-year-old Navigation data-validation flaw patched in Chrome 145; crafted HTML could potentially let a compromised renderer trick the browser into reading local files.
- Google says its AI systems analyze source code at rest on locked-down machines without general internet access, and that AI also assists with triage, validation, and candidate fixes.
- The company is piloting twice-weekly security releases and working on dynamic patching and other ways to shorten the time between fixes and updates reaching users.
- Google described efforts to reduce memory-safety bugs, including expanding MiraclePtr, deploying MiracleObject, and moving more code to Rust.
Article Details
- Event Type
- Google reported using an AI agent harness to identify Chrome vulnerabilities and accelerate security fixes.
- Impact
- Google said more than 1,800 Chrome bugs were patched this year, including 1,072 security defects across Chrome 149 and 150. One AI-discovered flaw, CVE-2026-3545, was a 13-year-old sandbox escape that could have allowed a compromised renderer to trick the browser into reading local files; Google patched it in Chrome 145.