Researcher Claims Claude Opus Enabled Creation of Working Chrome Exploit

A security researcher used Anthropic's Claude Opus to help develop a working browser exploit chain for CVE-2026-5873 (a V8 out-of-bounds R/W fixed in Chrome 147) against an older Chromium build used by Discord Desktop. The experiment produced a proof-of-concept that achieved code execution on ARM64 macOS (launching Calculator), consumed ~2.33 billion tokens over 1,765 requests, cost ~$2,283 in API usage, and required ~20 hours of human supervision, highlighting that generative AI can materially speed up converting known browser bugs into practical exploits—especially against unpatched, bundled Chromium instances.