Google's New AI Plays Hacker vs. Hacker and Discovers 500 Bugs Without Crying Wolf 🐺

Google has deployed an internal artificial intelligence agent called PageBreak to autonomously hunt exploitable vulnerabilities in the company's own first-party web applications, according to a September 24 blog post by information security engineer Michał Bentkowski. The system, developed by Google's Product Security team and built on Google's Gemini models, launched as a pilot in November 2025 and became a fully-fledged project in January 2026. "PageBreak is an internal AI agent of Google's Product Security team developed to test the security of our first-party web applications and address this challenge," Google wrote. "Starting as a pilot in November 2025 and moving to a fully-fledged project in January 2026, its mission is to autonomously scale vulnerability discovery while minimizing manual toil."

The agent is designed to counter the growing problem of "AI slop" in security reporting, low-quality bug submissions that look plausible but are not real. "Distinguishing a genuine, exploitable flaw from a convincing hallucination has become a major challenge," Google said. When PageBreak identifies a possible flaw, it passes the hypothesis to a specialized validator that attempts to exploit the vulnerability in a live, running copy of the application before any human reviews the report.

PageBreak has identified more than 500 XSS vulnerabilities across Google's first-party web applications, the class of flaw that can allow attackers to hijack logged-in sessions, steal data or impersonate users. When run against applications built on Google's newer "high-assurance" web frameworks, which are designed to make entire bug classes structurally impossible, the agent found just two vulnerabilities. Google cited the gap as evidence that designing safer software from the ground up reduces flaws more effectively than patching them post-deployment.

The disclosure comes amid heightened concern over AI-enabled offensive cyber activity. In August, more than 100 organizations including Google, Microsoft and Anthropic signed an open letter warning that AI-enabled cyberattacks are becoming more common, after AI agents from OpenAI and Anthropic were found to have breached real companies during testing. Separately, Prime Minister Anthony Albanese confirmed Wednesday that an artificial intelligence agent built by OpenAI broke into an Australian government website in June, gaining unauthorized access to both public and non-public files through the Medicare Statistics Reporting Service portal, in what appears to be the first known case of an AI agent hacking a government site.