Security startup XBOW has published details of a Chrome full-chain exploit that it says affected an estimated one billion users, revealing that the chain was built in just six days with the help of OpenAI's publicly available GPT-5.6 Sol model — and earned the company a $250,000 bounty from Google, the first awarded under the company's 2026 vulnerability reward rules. The disclosure appeared in a company blog post on October 13 and was picked up by The Hacker News the following day.
A full chain is the most valuable class of browser bug: a set of vulnerabilities that, chained together, let an attacker go from an ordinary web page all the way to executing code on the underlying machine. In this case, XBOW says the chain showed how simply visiting a website in Chrome could allow a remote attacker to take over a Windows computer. All of the bugs involved have now been fixed, and the company published its write-up after the patches landed. For readers tracking how quickly AI capabilities are changing security research, our breaking AI news coverage follows these developments as they happen.
Six Days Instead of Six Months
The most striking number in the disclosure is not the bounty but the timeline. XBOW says an intelligence-agency veteran it consulted estimated that his former team would once have needed around six months to identify this kind of vulnerability and achieve total access to a target. The company's four-person team, working alongside GPT-5.6 Sol and its predecessors, produced the awarded chain in six days.
The researchers — Brendan Dolan-Gavitt, Andres Luksenberg, Baul Lee, and Willem Pinckaers — deliberately chose a model that anyone can access, rather than a specialized security system or one with restricted availability. XBOW says the compute used over the six days of work on the awarded chain cost roughly $40,000, a fraction of the $250,000 bounty the chain ultimately won from Google.
According to the blog post, the division of labor was clear. The AI searched Chrome's source code, tested ideas, built proofs of concept, and wrote early drafts of the exploit code. The human researchers set strategy, chose where to look, checked the agents' output, and redirected them based on results. XBOW notes that one human decision in particular proved essential to the final chain.
A Disqualified First Attempt and a Las Vegas Rebuild
The path to the bounty was not linear. XBOW set the Chrome full-chain as an internal challenge at the start of the year, targeting the Black Hat security conference in August as its deadline. The team's first complete chain was disqualified after Google informed the company that it already knew about one of the bugs in the set.
Because the team had deliberately structured the chain as three independent parts, each with an alternative bug available, it could rebuild quickly. Working from Las Vegas during the week of Black Hat and DEF CON — between Sunday, August 2 and Friday, August 7 — the researchers and their agents assembled a fresh chain drawing on the earlier attempt. That second chain qualified, winning the $250,000 first prize under Google's updated 2026 rules.
What It Means for AI and Offensive Security
The disclosure lands at a moment when the security industry is actively debating whether AI systems meaningfully change the economics of vulnerability discovery. XBOW's argument is that they do, and dramatically: the company says it believes that within a year, each of the three bugs in its chain could plausibly be found by autonomous agents working without human handling.
There are caveats worth stating. The $40,000 compute bill is far from trivial for most attackers, and the work still required experienced human researchers to steer the process, validate findings, and make the judgment calls that turned raw agent output into a qualifying chain. XBOW itself frames the result as a partnership between human expertise and machine speed, not a demonstration of fully autonomous exploitation.
Still, the benchmark shift is hard to ignore. A class of exploit that an elite government team would budget six months for was replicated by a four-person startup with a commercial AI model, a cloud compute budget smaller than the reward check, and less than a week of calendar time. Chrome is used by more than half of the world's internet users, which is precisely why Google invests so heavily in hardening it — and why the speed at which AI-assisted research can now probe those defenses matters well beyond the security community.
For defenders, the XBOW post is double-edged. The same agent-driven workflow that found the bugs is available to the blue teams and browser vendors racing to fix them, and Google's bounty program — which paid out promptly under its new rules — is part of the machinery that got the flaws patched before public disclosure. The full technical details of the three bugs are available in a companion post by XBOW researcher Brendan Dolan-Gavitt.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
