Google says artificial intelligence is helping its security teams discover and fix Chrome vulnerabilities at a significantly faster rate. More than 1,000 security bugs were patched across the browser’s two latest releases as the company expanded its use of AI throughout the vulnerability management process.
According to Google, Chrome 149 and Chrome 150 addressed a combined 1,072 security vulnerabilities. That figure exceeds the total number of security bugs fixed across the previous 23 Chrome milestones combined.
Google now uses large language models (LLMs) across multiple stages of Chrome vulnerability management. These tasks include finding security flaws, reproducing vulnerability reports, assessing severity, assigning bugs to developers, generating potential patches, and creating security tests.
Google began using LLMs to improve security fuzzing in 2023. The company later worked with Project Zero on Naptime, an AI system that gives models specialized tools for vulnerability research.
Google then partnered with Google DeepMind and Project Zero to develop Big Sleep, an AI-powered vulnerability discovery agent. Big Sleep identified security flaws in Chrome’s V8 JavaScript engine and graphics components.
In early 2026, Google introduced a Gemini-powered agent harness designed to scan the broader Chrome codebase for vulnerabilities while reducing false-positive reports.
Source: Google
One vulnerability discovered by Google’s AI systems was a Chrome sandbox escape that had remained in the codebase for more than 13 years. If exploited, the flaw could have allowed a compromised renderer to escape the browser sandbox and access local files.
Google is also encouraging Chrome developers to create SECURITY.md files that document trust boundaries and threat models. This information helps AI systems identify code and operations with potential security implications.
The company says its multi-agent AI workflows complement, rather than replace, established security testing techniques. Traditional fuzzing remains an important method for uncovering complex Chrome vulnerabilities.
Google has also reported a sharp increase in submissions to the Chrome Vulnerability Reward Program. By March 2026, the company had received more security bug reports than during all of 2025.
In response, Google modified the program to prioritize vulnerability reports that provide information beyond what its automated security tools are already finding and processing.
AI is also being used to automate vulnerability triage. The process includes filtering spam and duplicate reports, reproducing proof-of-concept exploits, assigning severity ratings, and routing confirmed issues to the appropriate developers.
Google estimates that its automated vulnerability management process saves hundreds of developer hours every month.
After a security flaw is confirmed, AI fixing agents generate multiple potential patches. Another agent reviews the proposed fixes and provides additional technical information for developers to evaluate.
In May, Google said these systems helped prevent more than 20 vulnerabilities from reaching production, including one issue classified as critical.
However, Google says faster vulnerability discovery and remediation must be matched by faster Chrome security updates.
Once a security fix is added to Chrome’s public source code, attackers can analyze the change and attempt to reverse-engineer the vulnerability before the update reaches users.
To reduce this patch gap, Google is moving Chrome toward a two-week major release cycle with weekly security updates. The company is also testing two security releases per week.
Google is developing “dynamic patching” to reduce disruption by allowing Chrome to apply certain updates without requiring a browser restart.
Beginning with Chrome 150 on macOS, Chrome can automatically restart to install a pending update when the browser is running in the background and no windows are open.
Google’s long-term goal is to keep Chrome continuously updated through dynamic patching, automatic restarts during periods of inactivity, and improved session restoration.
Security teams log 54% of successful attacks and alert on just 14%. The remaining threats can move through your environment undetected.
The Picus whitepaper explains how breach and attack simulation can test SIEM and EDR detection rules, helping security teams identify gaps before attackers exploit them.
Source: www.bleepingcomputer.com