Making AI Headlines · Page 3

Google's Bug Hunter Finds 500 Reasons Its Human Colleagues Cannot Take Friday Off

October 31, 2026 at 6:01 AM EDT · New York time

Google says its PageBreak AI security agent has found more than 500 cross-site scripting vulnerabilities across its web applications since entering full operation in January. The system uses Gemini models to investigate possible flaws, according to a Google paper, and reports them only after deterministic validators reproduce the exploit against a running application. Cross-site scripting can allow malicious code to run through a trusted web page. Finding a flaw is therefore useful, but a finding still needs people and processes to decide what to fix and how to deploy the repair. The validation step is important too: a security team with hundreds of real problems has little reason to ask for hundreds of imaginary ones. PageBreak turns some of the search into a machine task, while leaving a substantial list of follow-up work. The system's success is also a mildly uncomfortable tour of Google's own property. The company built an agent to look for weaknesses and discovered that the agent would have plenty to do. Automated security may make the search faster, but it does not make a vulnerability disappear when the report is printed. For the engineers receiving the results, the robot is not replacing the to-do list. It is a remarkably productive person adding items to it.

Satire. Not factual reporting. Read the original reporting from the linked publishers.