Xbow USA Finds Chrome Exploit Chain With $40K AI Compute

Researchers at Xbow USA spent just $40,000 in AI compute time over six days to uncover a chain of vulnerabilities in Google’s Chrome browser that could let attackers take control of up to one billion Windows machines worldwide. The flaw involved an escape from the sandbox that contains the V8 JavaScript engine and a second escape from Chrome’s operating‑system process sandbox, allowing a malicious web page to launch a Windows application. Google awarded the team a $250,000 bug bounty, the first payout under the company’s new 2026 reward rules.

How the exploit worked

The team used the AI model GPT‑5.6 Sol to search Chrome’s source code, test attack ideas, and build proofs of concept. While the model suggested code snippets and potential attack vectors, the researchers decided on the overall strategy, selected target areas, and verified the model’s output. GPT‑5.6 Sol helped them identify the precise conditions that allowed the sandbox escapes and guided the construction of the final exploit.

After the researchers reported the flaw to Google, the company released patches to close the vulnerabilities. Google confirmed that the fix would prevent remote code execution from compromised web pages on Windows systems. Xbow USA made all details of the exploit and the development process available in a public blog post, highlighting how inexpensive AI compute can accelerate security research.