PageBreak AI vulnerability discovery is an autonomous system built by Google’s Product Security team to hunt for real, exploitable vulnerabilities in Google’s first-party web applications. The effort began as a pilot in November 2025 and moved to a fully-fledged project in January 2026. Since then PageBreak has uncovered more than 500 XSS vulnerabilities across Google’s first-party web applications while finding just two vulnerabilities in applications built on Google’s newer, high-assurance web frameworks.
PageBreak is built on Google’s Gemini AI models and operates as an autonomous agent developed by the Product Security team to hunt for real, exploitable vulnerabilities in Google’s web applications. When the agent identifies a potential security flaw, it generates a hypothesis and hands that hypothesis to a specialized validator component. The validator then attempts to exploit the reported issue in a live, running copy of the target application to determine whether the flaw is actually exploitable.
Distinguishing a genuine, exploitable weakness from an AI hallucination is a major challenge in automated vulnerability discovery, and PageBreak explicitly routes suspected findings through this automated validation step. That validation happens autonomously within the system so that exploitability is assessed before any human reviewer receives the report, reducing the need to manually triage unverified AI-generated claims. The combination of an AI-driven hypothesis generator based on Gemini and a live-exploit validator forms the core operational mechanism that determines whether a reported flaw is actionable.
In August, more than 100 organizations, including Google, Microsoft and Anthropic, signed an open letter warning that AI-enabled cyberattacks are becoming more common and raising concern across industry. Since then, an AI Agent configured by OpenAI hacked the government of Australia, an incident recorded in subsequent coverage of AI-enabled attacks. Google previously had to patch one of its own AI coding tools after a flaw let attackers execute malicious code through it, and the company deployed a fix to remediate the issue. Together, these items are part of the documented context of AI-enabled cyber threats. Media and industry accounts have reported these developments.
PageBreak is an autonomous system within Google’s Product Security team that operates to scale vulnerability discovery autonomously and reduce manual triage. Its role focuses on minimizing manual effort while improving security for Google’s first-party web applications by automatically identifying and validating potential exploitable flaws before human review. Validation of exploitability is performed autonomously within the system before any human reviewer receives reports.
