Google PageBreak

PageBreak is Google's AI agent for web application security testing, designed to solve the false-positive problem in AI vulnerability discovery. Rather than dumping hundreds of questionable findings on security teams, it selects validation logic appropriate to each suspected weakness and technology stack, then deterministically proves exploitability as part of the discovery pipeline. Google reported it finding 500+ web app flaws with exploit validation built in. It primarily uses Gemini models (3.1 Pro, 3.5 Flash) but the architecture is model-agnostic. Google intends PageBreak and CodeMender to work together as a pipeline: PageBreak discovers and proves, CodeMender patches, engineers review — an increasingly autonomous find-prove-fix loop.

The honest picture: PageBreak is the discovery half of a pipeline whose remediation half (CodeMender) is still in preview, so end-to-end autonomy isn't here yet. Details on availability and pricing are limited since it's part of Google's enterprise security motion. Deterministic exploit validation is powerful but can't cover every vulnerability class, particularly logic flaws that require business-context understanding.

Available through Google's enterprise security offerings. Contact Google Cloud for access details as part of the AI Threat Defense ecosystem.