Google describes a security agent that validates findings before escalation.
Reproducing suspected vulnerabilities helps separate useful reports from unverified agent output.
What happened
In a September technical post, Google reports that PageBreak uncovered more than 500 cross-site scripting vulnerabilities in its own web applications. Google says non-AI-written validators reproduce suspected exploits against running applications before findings are escalated. The count is Google-reported, not independently audited here. This is background carried forward following October coverage, not a new November launch.
Why it matters: Reproducing suspected vulnerabilities helps separate useful reports from unverified agent output.
Source: Google, “Agentic Hacks, Real Proofs: Inside Google's PageBreak Project”