AI Model Reportedly Finds V8 Zero-Day in 13 Minutes for $4.24
Vulnerability Disclosure · 2 sources · Updated 14h ago · First seen Nov 16, 2026

Xint.io reportedly found that an AI model could identify a Google V8 vulnerability in 13 minutes at a usage cost of $4.24, highlighting falling barriers to AI-assisted vulnerability discovery. In a restricted source-code test involving seven models, only GPT 6 Sol, GPT 5.6 Sol, and Claude Opus 5 identified the privately reported type-confusion flaw, identified in the reports as CVE-2026-102321, which could enable remote code execution. A February 2026 attempt had failed after 25 hours and $350, while GPT 6 Sol succeeded in September using a guided prompt.
The results suggest that improving models and declining inference costs could make vulnerability discovery more accessible to attackers and defenders, but they do not demonstrate autonomous discovery across arbitrary software: the tests used restricted source code, expert guidance remained important, and some attempts required additional prompting. The reports establish neither active exploitation nor patch availability. Security teams should evaluate AI-assisted testing within authorized workflows while maintaining layered defenses, including hardening, zero-trust access controls, network segmentation, and endpoint monitoring.


