XBOW's AI agents built a $250,000 Chrome exploit chain
Security firm XBOW paired researchers with OpenAI's GPT-5.6 Sol to chain three Chrome bugs into a working exploit in six days, earning a bounty for a flaw Google says reaches about a billion users.
XBOW disclosed on 13 October that it built a full exploit chain for Chrome by pairing its researchers with AI agents running OpenAI’s GPT-5.6 Sol, and that Google paid out a $250,000 bounty for it. The chain links three separate bugs: a Maglev compiler flaw in V8 that lets two JavaScript objects share a mutable number during deoptimisation, a WebP decoder bug that reads stale frame data to write out of bounds, and a Crashpad flaw on Windows that corrupts memory during a retried read. Together they take an attacker from JavaScript in a tab to code execution outside the renderer sandbox. Google has fixed all three.
The team built the exploit over six days in August, spending around $40,000 on compute. XBOW says the AI agents searched Chrome’s source, tested ideas and wrote early exploit drafts, while humans set strategy, chose targets and checked the results. Google estimates the flaw reached roughly a billion Chrome users on Windows.
Why it matters: full-chain browser exploits have traditionally taken skilled teams months to assemble. A security firm doing it in under a week, even with heavy human steering, changes the threat model for anyone who assumed that timeline bought them breathing room to patch.
The caveat: XBOW is explicit that humans picked targets, redirected the agents and verified every result. This is not evidence of autonomous exploit discovery.