Brendan Dolan-Gavitt, Andres Luksenberg, Baul Lee and Willem Pinckaers worked with AI agents to search Chrome’s source code, test ideas and build proofs of concept for an exploit chain.
Between August 2 and August 7, the XBOW team developed a chain that achieved code execution outside Chrome’s renderer sandbox on Windows. Google awarded XBOW $250,000 for the discovery.
The researchers demonstrated the result by launching Windows Calculator from a webpage using signed Chrome Beta.
The work began with Maglev, V8’s mid-tier optimizing compiler. It then targeted animated WebP decoding through WebCodecs ImageDecoder, before using Crashpad, Chrome’s crash-reporting handler, which operates outside the renderer sandbox.
Advertisement
XBOW used OpenAI’s publicly available GPT-5.6 Sol model and its predecessors throughout the work, using them to inspect code, test hypotheses, create proofs of concept and develop stages of the exploit. The team estimated API charges of around $40,000 for the six-day campaign.
The payout was Google’s first for a full-chain Chrome exploit under a vulnerability reward programme introduced in April 2026. The programme offers financial rewards for researchers who find and report chains of flaws capable of escaping Chrome’s security sandbox.
According to XBOW, all three vulnerabilities used in the chain have since been fixed. The changes prevent reuse of mutable HeapNumbers, keep WebP output buffers within actual bitmap boundaries, and limit Crashpad retries to the original requested length.
The discovery arrives alongside a wider run of Chrome security work. Chrome 154 fixes 108 security flaws, including 11 critical bugs, and brings security fixes without new features. Multiple Google Chrome vulnerabilities were dated 2026-10-13 and 2026-10-14.
Advertisement
Comments (0)