Google Deploys PageBreak AI Agent to Hunt Security Flaws

Google revealed on September 24 that its Product Security team developed an autonomous system named PageBreak. The tool is designed to search for real and exploitable vulnerabilities inside Google web applications. Information security engineer Michał Bentkowski detailed the project in a blog post.

PageBreak began as a pilot in November 2025 before transitioning into a fully active project in January 2026. Its primary mission is to scale vulnerability discovery autonomously while reducing manual workloads.

Security teams across the industry have struggled with low-quality, AI-generated bug reports. Google noted that these false reports look plausible but waste valuable time. Distinguishing a genuine flaw from a hallucination has become a major challenge for security professionals.

Google built PageBreak using its Gemini models. When the agent detects a potential security flaw, it sends the hypothesis to a specialized validator. This validator attempts to exploit the issue inside a live copy of the application.

PageBreak has already discovered more than 500 cross-site scripting vulnerabilities across Google first-party web applications. These types of flaws can allow attackers to hijack logged-in sessions, steal data, or impersonate users.

Google tested the agent against applications built on newer, high-assurance web frameworks intended to eliminate entire bug classes. PageBreak found only two vulnerabilities in those applications. Google stated this disparity shows that building safer software initially works better than patching holes later.

The broader cybersecurity landscape has seen increasing integration of artificial intelligence. More than 100 organizations, including Google, Microsoft, and Anthropic, signed an open letter in August warning about rising AI-enabled cyberattacks. That warning followed incidents where AI agents from OpenAI and Anthropic breached real companies during testing. An OpenAI agent also hacked the government of Australia.

PageBreak represents the defensive side of this technology by using artificial intelligence to catch bugs before malicious actors do. Google previously patched one of its own AI coding tools after a flaw allowed attackers to execute malicious code.

Google stated that PageBreak relies on internal advantages that most companies lack. These advantages include a unified code repository containing billions of lines and years of internal scanning infrastructure, meaning small startups cannot easily replicate the method.

The next known step involves connecting PageBreak to CodeMender, Google's automated patch-writing agent. This connection will allow confirmed vulnerabilities to arrive with proposed fixes already attached for engineers to review and approve.