Xbow built a 1B-user Chrome exploit in 6 days on GPT-5.6 Sol, and Google paid $250,000

#What happened

Security startup Xbow disclosed on October 13 that its AI agents, running OpenAI's publicly available GPT-5.6 Sol model, helped chain three previously unknown Chrome vulnerabilities into a full-chain exploit. The chain combined a V8 Maglev compiler bug, an animated-WebP decoder bug, and a Crashpad overflow to cross both of Chrome's sandboxes and run code outside the browser on Windows machines, with no click or download needed. Xbow estimated about one billion Chrome-on-Windows users were in the path of the flaw. Google confirmed all three bugs, fixed them (in Chrome 152.0.7977.65 on August 25 and 152.0.7977.82 on September 3), and paid Xbow $250,000, its top reward for a Chrome full-chain exploit and the first full-chain award under Google's 2026 rules. The six-day sprint ran August 2 to 7 during the Black Hat and DEF CON conferences and cost about $40,000 in AI compute. One security researcher directed the agents while human researchers Brendan Dolan-Gavitt, Andres Luksenberg, Baul Lee, and Willem Pinckaers chose targets, tested hypotheses, and decided which components to chain. CEO Oege de Moor said an intelligence-agency veteran told him his team used to spend six months on this kind of vulnerability; Xbow did it in six days.

#Why it matters

This is a concrete demonstration of publicly available AI models collapsing the cost and time of finding and chaining real vulnerabilities. A chain that previously took a team months was built by one researcher plus agents in under a week for roughly $40,000 of compute, and it earned a $250,000 payout. The security researcher who wrote up the chain concluded "the bar is significantly lower" for attackers armed with LLMs. Practical note: Chrome updates itself, but the fix only takes effect after a relaunch, so check that you are on 152.0.7977.82 or later.

Verified October 15, 2026.