AN AI HACKED EVERY LAYER OF CHROME IN SIX DAYS

Researchers at XBOW, working with AI agents running OpenAI's publicly available GPT-5.6 Sol model, built a full-chain exploit for Google Chrome in six days. The chain linked three previously unknown vulnerabilities, starting in the V8 JavaScript engine, escaping its sandbox through a flaw in animated WebP image processing, and then using a bug in Chrome's Crashpad handler to achieve remote code execution on Windows simply by loading a web page.

XBOW said humans set strategy, chose where to look and checked the model's work while the AI searched Chrome's source code, tested ideas and built proofs of concept. Google fixed all three flaws. The company presented the result as evidence of how quickly a public frontier model can collapse exploit-development timelines that once required elite teams.