XBOW’s $250,000 Sol Chrome Exploit Took Six Days to Build

In reportedly early 2026, xbow set a challenge in Malta that would eventually lead to a reportedly $250,000 Chrome sandbox escape. The exploit chain took just six days to develop, putting long-held assumptions about browser security to the test. Details about the project come from xbow’s own account and reporting, not independent verification.

early 2026: xbow sets the Chrome challenge

At a company meeting in Malta in early 2026, xbow founder and CEO Oege de Moor set a goal: build a complete Chrome exploit chain. Four researchers took on the work: Brendan Dolan-Gavitt, Andrés Luksenberg, Baul Lee and Willem Pinckaers.

The goal was a full chain: a series of vulnerabilities that could work together to cross more than one security boundary. It was a substantial challenge. In a technical write-up, Dolan-Gavitt compared the effort with the team’s earlier work, saying a four-person team would previously have needed six to nine months to produce a full Chrome chain. XBOW’s first submission took two months, but didn’t earn the award because Google said it already knew about one of the bugs.

August 2–7: AI agents help build the awarded chain

The successful attempt ran from August 2 to August 7, while XBOW’s team was in Las Vegas for Black Hat and DEF CON. According to reports from The Next Web, XBOW used AI agents running a model identified as GPT-5.6 Sol to help create the exploit. The company put compute costs at about $40,000, based on its usage logs.

The chain joined three previously unknown Chrome bugs. XBOW described an attack in which simply loading a web page could let code run outside Chrome’s sandbox on a Windows computer, without a click or download. The sandbox is designed to limit what a compromised page can do, so escaping it could increase the harm a malicious site causes. The six-day figure covers the awarded chain, not the entire project: the challenge began months earlier, and XBOW’s first submission had already taken two months. De Moor said the company deliberately chose a model available to anyone, framing the work as a test of what AI agents could do with accessible tools.

October 13: Google’s $250,000 award becomes public

On October 13, XBOW announced it had received $250,000 through Google’s Chrome bug bounty programme. The company described the reward as the first full-chain Chrome award under rules Google introduced in April 2026.

The bounty is the headline figure; XBOW put its compute expense at about $40,000. According to the company, Google fixed all three vulnerabilities, and none had been used against Chrome users. Before the fixes, XBOW estimated that roughly one billion people using Chrome on Windows could have been exposed. That figure is the company’s estimate of potential reach, not a confirmed count of affected or attacked users.

XBOW says the exploit chain earned a $250,000 bounty after six days of development, with about $40,000 in compute costs.

What the sol chrome exploit signals next

The episode challenges a familiar assumption about browser security: that advanced exploit development must be slow and limited to specialist teams. XBOW’s account suggests AI agents can help researchers work faster, but it doesn’t show that the model discovered and assembled the chain on its own, without human direction.

Four named researchers worked on the broader challenge, and months of preparation preceded the successful attempt. That distinction matters to defenders. Faster research can help security teams find and patch flaws, as Google did here, but it could also lower the time and resource barriers for attackers if similar capabilities are misused. Compute costs came to about $40,000, making the financial efficiency stand out. For future disclosures, the key question isn’t just how quickly an exploit was built, but how much of the work AI agents handled.