XBOW Wins $250,000 Chrome Bounty for AI-Built Exploit Chain
Security firm XBOW used AI agents to build a full Chrome exploit chain in six days, earning a $250,000 bounty from Google.
Key points
- XBOW chained three unknown Chrome bugs to execute code outside the sandbox on Windows.
- The exploit was built using OpenAI’s GPT-5.6 Sol model in six days at a compute cost of $40,000.
- Google patched the flaws, with Chrome 152 release notes crediting the researcher for CVE-2026-78952.
Security company XBOW announced on October 13 that it received a $250,000 bounty from Google for discovering and building a full exploit chain for the Chrome browser. The firm used artificial intelligence agents to identify and chain three previously unknown vulnerabilities, significantly reducing the time typically required for such research.
XBOW researcher Brendan Dolan-Gavitt stated that the awarded exploit chain was constructed between August 2 and August 7. The team utilized OpenAI’s GPT-5.6 Sol model, incurring approximately $40,000 in compute costs. This six-day effort followed months of preparation and represents a shift from traditional manual vulnerability discovery methods.
Technical Context and Development
The exploit chain targeted Chrome on Windows, allowing code execution outside the browser’s sandbox simply by loading a web page. No user clicks or downloads were required. The chain began with a flaw in the V8 engine’s Maglev compiler, which provided read and write access within the V8 sandbox.
A second vulnerability in the animated WebP image decoder allowed a single write operation outside the V8 sandbox. The third flaw, located in Chrome’s Crashpad crash handler, converted this write into full code execution. Google’s release notes for Chrome 152 credit Dolan-Gavitt of XBOW for the Crashpad flaw, CVE-2026-78952, rated high.
XBOW noted this is the first full-chain Chrome award under rules Google introduced in April 2026. The team, including Andrés Luksenberg, Baul Lee, and Willem Pinckaers, had previously submitted a chain that failed because Google already knew one of the bugs. They swapped in a backup stage to create the successful submission.
Impact and Patch Status
XBOW estimates that the unpatched bugs exposed approximately one billion people running Chrome on Windows. The company stated that the vulnerabilities were never used against Chrome users before the fixes were released. Google fixed all three bugs, with updates shipping on August 25 and September 3, according to reporting by The Hacker News.
The rapid development of the exploit, compared to the six to nine months typically required by a team of four before large language models, highlights the changing landscape of vulnerability research. As Threat Times reported on 2026-10-16, Xbow researchers used AI to find this Chrome exploit chain, earning the substantial bounty.
What should you do?
Users should ensure their Chrome browsers are updated to the latest version to receive the security patches for CVE-2026-78952 and the associated vulnerabilities. Organizations should verify that automated update mechanisms are functioning correctly across all Windows endpoints running Chrome.
Background: Large language models
Large language models are a specific type of neural network trained on text. Generative AI is the broader umbrella term for any system that creates new content, including images and code. Not all generative AI uses large language models, but most modern text generators do.
What to do and how to stay safe: Chrome
- Verify that all Windows devices are running the latest Chrome updates to ensure the Crashpad and V8 fixes are applied.
- Monitor internal logs for unusual browser crashes or unexpected processes launching from the Chrome directory.
- Restrict access to untrusted web content where possible to reduce the attack surface for drive-by exploits.
- Review browser sandbox configurations to ensure they are enabled and functioning as intended.
Step-by-step guide: Mobile Device Security: How to Protect Data Beyond Passwords
General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
What CVE ID is associated with the Crashpad vulnerability?
The Crashpad vulnerability is identified as CVE-2026-78952.
How much did it cost XBOW to build the exploit?
XBOW estimated the compute cost at about $40,000.
Was the exploit actively used in the wild?
XBOW states the bugs were never used against Chrome users before the fixes were released.
