Researchers Used OpenAI Model to Build a Chrome Exploit Chain in Six Days

What the Researchers Did

According to The Hacker News↗ (opens in a new tab), a team at security firm XBOW managed to chain together three previously unknown bugs in Google Chrome. This allowed them to create a full attack chain that could run code on a Windows computer just by loading a web page. The process took only six days, with the help of AI agents running OpenAI's GPT-5.6 Sol model.

The team built the exploit during the Black Hat and DEF CON conferences in August 2026. They had already spent months preparing, but the final chain was assembled in less than a week. Google confirmed the bugs, fixed them, and paid XBOW its top reward of $250,000 for a full-chain Chrome exploit.

How the Exploit Chain Worked

A full-chain exploit is a series of vulnerabilities used together to break through all the security layers of a program. In this case, Chrome's protections include two sandboxes, barriers that keep web code from reaching the rest of the computer. The exploit chain crossed both sandboxes and ended up running code outside Chrome, on the Windows system. The user did not need to click or download anything, just loading a web page was enough.

The three bugs were in different parts of Chrome. One was in the V8 JavaScript engine, another in the code for reading animated WebP images, and the third in Crashpad, Chrome's crash handler. The second bug let the attacker break out of the inner sandbox, and the third allowed full code execution on the machine.

The Role of AI in the Attack

XBOW used OpenAI's GPT-5.6 Sol model, which is publicly available. The AI agents searched Chrome's source code, tested ideas, built proofs of concept, and wrote first drafts of the exploit. Human researchers chose the targets, provided context, and decided how to join the pieces. According to XBOW, the six-day run cost about $40,000 in AI compute charges.

XBOW's Brendan Dolan-Gavitt said that before large language models (LLMs), building a Chrome full-chain exploit would take a team of four people around six to nine months. Their first attempt with AI took two months. The awarded exploit took just six days.

What Google Did and What Users Should Do

Google confirmed and fixed all three bugs. Two of them were patched in Chrome version 152.0.7977.65 (August 25) and 152.0.7977.82 (September 3). Chrome updates itself, but the fix only takes effect after restarting the browser. Users should relaunch Chrome and check that their version is 152.0.7977.82 or later.

XBOW and Google both say the bugs were not used against Chrome users before they were fixed. XBOW calls this the first full-chain Chrome reward under Google's 2026 rules, and the first to an autonomous security company.

The Bigger Picture: Speed and Cost

XBOW's main point is the speed and cost. It took only six days and $40,000 to build a chain that could have put an estimated 1 billion Chrome users at risk. The company warns that the bar for building these kinds of attacks is now much lower. Defenders need to adapt to this new reality, where AI can help attackers move much faster than before.

What You Should Do

• Restart Chrome to make sure you have the latest version.
• Stay aware of browser updates and apply them promptly.
• Security teams should review their defences, as AI is making advanced attacks faster and cheaper to build.

What Is Still Unclear

This story comes from a single report by The Hacker News. There is no primary source or confirmation from Google, Microsoft, or OpenAI in the public domain as of now. Some technical details, such as the exact identifiers for the bugs, are still unclear.