Google Earns $250K Bounty for Critical Chrome Exploit Found with GPT-5.6

Key Takeaways

• Security researchers at XBOW successfully developed a full-chain exploit for Google Chrome using AI agents, earning a $250,000 bounty.
• The exploit chained three distinct vulnerabilities (CVEs not provided in source) to achieve code execution outside Chrome’s sandbox.
• The attack specifically targeted Windows systems and demonstrated the potential to affect up to one billion users globally.
• Google has since patched all three identified flaws, reinforcing the importance of keeping Chrome updated.

XBOW Researchers Leverage AI to Craft Critical Chrome Exploit, Secure $250K Google Bounty

Security researchers at XBOW have been awarded a substantial $250,000 bounty by Google for constructing a sophisticated full-chain exploit targeting the Chrome browser. This groundbreaking achievement was notable for its reliance on AI agents, specifically OpenAI’s publicly available GPT-5.6 Sol model, to facilitate the exploit’s development.

The entire endeavor, from initial research to a functional exploit, reportedly took just six days and incurred approximately $40,000 in computing costs. Google has confirmed that the underlying vulnerabilities have been addressed, and users are urged to ensure their Chrome browsers are updated.

Unpacking the Full-Chain Attack

XBOW publicly disclosed the details of their research on October 13, 2026. Their full-chain exploit meticulously linked three separate bugs, allowing an attacker to escalate privileges from executing JavaScript within a Chrome tab to achieving code execution outside the browser’s renderer sandbox. While the demonstration specifically targeted Windows, XBOW estimated the potential impact could extend to approximately one billion users worldwide.

A browser sandbox is a critical security mechanism designed to isolate web content and limit the damage a compromised webpage can inflict. A full-chain exploit systematically bypasses these defenses. XBOW’s proof-of-concept involved a specially crafted web page that successfully launched an arbitrary program on a test system, illustrating the severity of the chained vulnerabilities. The company emphasized that there was no evidence of this exploit being used maliciously against users in the wild.

The Three Vulnerabilities

The exploit comprised a sequence of three distinct flaws:

1. V8 JavaScript Engine Vulnerability

The initial vulnerability resided within V8, Chrome’s high-performance JavaScript engine. XBOW identified a bug in the Maglev compiler where, during the deoptimization process (when V8 transitions from optimized to less optimized code), separate JavaScript objects could inadvertently share a mutable number. Manipulating this shared value allowed the exploit to confuse optimized code, leading to unauthorized memory access within V8’s protected memory regions. More technical details can be found in XBOW’s research: Zero-to-Calc in Six Days: Inside XBOW Chrome Full-Chain.

2. Animated WebP Image Decoder Flaw

The second vulnerability involved the animated WebP image decoder. Researchers discovered that the decoder could erroneously reuse frame details from one image while processing altered image data. This mismatch resulted in an out-of-bounds write, enabling the exploit to escape the V8 sandbox and progress further into the system.

3. Crashpad Crash-Reporting System Bug

The final vulnerability exploited a flaw in Crashpad, Chrome’s crash-reporting system, specifically on Windows. Crashpad, when retrying a failed memory read, could calculate the length of the retry based on the remainder of a page. Crucially, the code assumed that the renderer’s state would remain unchanged during this retry. XBOW’s exploit manipulated page access permissions during the retry operation, leading to an oversized read and subsequent memory corruption within the Crashpad handler, which operates outside the renderer sandbox. Additional insights are available in their detailed write-up: Inside XBOW Chrome Full-Chain.

AI’s Role and Human Oversight

XBOW researchers underscored that while AI agents significantly accelerated the process, human expertise remained paramount. The research team guided the AI, selecting targets, delegating tasks, meticulously reviewing AI-generated results, and preparing contingency plans. For instance, an early iteration of the exploit chain had to be replaced after Google informed XBOW that one of its components was already known. The modular design of their approach allowed them to seamlessly integrate an alternative bug. The company said it developed the final chain between August 2 and August 7.

Google’s Vulnerability Reward Program

Google’s Chrome Vulnerability Reward Program (VRP) offers up to $250,000 for eligible full-chain exploits that can be executed from web content on a current Chrome release. The program mandates a working demonstration and adherence to responsible testing guidelines. It’s important to note that the reward acknowledges the reported flaw and its exploitability, not confirmed instances of user device compromise. More information on the program can be found in Google’s reward rules.

This incident highlights the growing role of AI in cybersecurity research, potentially shortening the time required to uncover complex vulnerabilities. Software development teams are encouraged to leverage these tools for internal product testing, always ensuring human review and coordinated patch deployment remain central to their security strategies.

What You Should Do

• Update Google Chrome Immediately: Ensure your Chrome browser is running the latest version. Relaunch the browser when an update is available to apply patches.
• Enable Automatic Updates: Verify that Chrome’s automatic update feature is enabled to receive critical security fixes promptly.
• Exercise Caution with Untrusted Websites: While these vulnerabilities are patched, always be wary of visiting unfamiliar or suspicious websites.
• Keep Operating System Updated: Ensure your operating system (e.g., Windows) is also fully patched, as exploits often chain browser vulnerabilities with OS-level flaws.

Disclaimer: HackersRadar reports on cybersecurity threats and incidents for informational and awareness purposes only. We do not engage in hacking activities, data exfiltration, or the hosting or distribution of stolen or leaked information. All content is based on publicly available sources.