AI Agents Help Researchers Build Chrome Full-Chain Exploit in Six Days

· Original article ↗

Summary

XBOW says researchers used OpenAI’s public GPT-5.6 Sol model to build a Chrome full-chain exploit in six days. Google fixed all three bugs and awarded XBOW $250,000; XBOW says the bugs were not exploited against users.

Key points

  • The exploit chained three previously unknown Chrome bugs to run code on Windows when a user loaded a web page, without clicking or downloading anything.
  • XBOW says AI agents searched Chrome’s source, tested ideas, built proof-of-concept code and drafted the exploit; researchers selected targets and evaluated and combined the results.
  • The awarded chain was built from August 2 to 7, after months of earlier work; XBOW estimates the AI usage cost about $40,000.
  • Google confirmed and fixed all three bugs. Two identified bugs were fixed in Chrome versions 152.0.7977.65 and 152.0.7977.82.
  • Chrome updates automatically, but users need to relaunch the browser for the fix to take effect and confirm they are running version 152.0.7977.82 or later.
  • XBOW says the bugs were never used against Chrome users; before fixes shipped, the company estimated that about one billion Chrome users on Windows were at risk.
  • Google paid XBOW its maximum $250,000 reward for a full Chrome exploit; XBOW says it was the first such award under Google’s 2026 rules to an autonomous security company.

Article Details

Event Type
Security research disclosure of a Chrome full-chain exploit and vulnerability reward
Impact
The three-bug chain could run code outside Chrome on a Windows computer when a user loaded a web page, bypassing Chrome's sandbox protections. XBOW said the bugs were never used against Chrome users. Google fixed all three bugs; before the fixes, XBOW estimated that about 1 billion Chrome users on Windows were at risk.

CVE

People

Vendors

Products

Related Articles