AI Models Exploit Chrome V8 with High Success Rate, Google Fixes Over 1,000 Bugs

AI models have demonstrated a 73% success rate in exploiting Chrome's V8 JavaScript engine, according to David Brumley. However, the actual goal of hacking is to gain control of the target machine, where leading models achieved zero successful sandbox escapes. Google has reported fixing 1,072 security bugs across Chrome versions 149 and 150, utilizing AI to enhance vulnerability management. The AI systems assist in discovering flaws, generating patches, and automating triage processes. Notably, a long-standing vulnerability allowing sandbox escapes was identified and fixed. The use of AI in Chrome's security has led to a significant increase in reported vulnerabilities and expedited patch delivery. Despite advancements, the effectiveness of AI in achieving control-flow hijacks remains under scrutiny.

Key Points: • AI models exploit Chrome's V8 engine at a 73% rate but fail to achieve control. • Google fixed 1,072 security bugs in Chrome 149 and 150, surpassing previous milestones. • AI tools are enhancing vulnerability management but have limitations in achieving sandbox escapes.