errors.x7.al › Recent issues

XBOW built a Chrome full-chain exploit with a public AI model

Security firm XBOW says it discovered and built a full-chain exploit for Google Chrome — affecting an estimated billion users — with help from a publicly available OpenAI model, in six days. According to XBOW's blog post published on 13 October 2026, the bugs in the chain have been fixed and the work earned a $250,000 Google bounty. The firm frames the result as a sign of how quickly AI is changing vulnerability discovery.

What's happening

According to XBOW, a "full chain" is a set of bugs used together to open every door, from an ordinary web page to the computer itself. The firm says its chain showed how an attacker could use Chrome to remotely take over a Windows machine, and that it estimates the impact at roughly a billion users. The three bugs involved are reported to be fixed.

XBOW says the effort was a human-machine partnership: four researchers — Brendan Dolan-Gavitt, Andres Luksenberg, Baul Lee, and Willem Pinckaers — worked alongside "OpenAI's GPT-5.6 Sol and its predecessors." It says it deliberately chose a model available to anyone rather than a restricted security model, and that the compute used over the six days cost $40,000. The work reportedly won $250,000 from Google, described by XBOW as the first bounty under Google's 2026 rules.

One detail stands out in XBOW's account: an earlier version of the chain was disqualified because Google already knew about one of its bugs. The team had built the exploit as three independent parts, each with an alternative, so it could swap in a backup. XBOW says it rebuilt the chain between Sunday 2 August and Friday 7 August 2026, while at Black Hat and DEF CON. The Hacker News has also reported that researchers used a public OpenAI model in this work.

Who's affected

The direct subject is Google Chrome. XBOW notes Chrome holds a solid majority of global browser market share and that more than half the world's population relies on it, and the demonstrated chain targeted Windows machines. Because the three bugs are said to be fixed, the practical risk to users running a current, relaunched Chrome is reduced, though the wider lesson applies to anyone running browser or endpoint software that can be reached from a web page.

The broader audience is defenders and software vendors. XBOW argues the cost and time of finding serious vulnerabilities are falling fast for attackers and defenders alike, and that within a year each of the three bugs in this chain could be found by AI agents working on their own. That prospect is likely to matter for any team that ships internet-facing software.

What to do

• Keep your Chrome browser up to date. XBOW says all three bugs in its chain are fixed and that Google ships fixes quickly — but a fix only takes effect once you relaunch the browser, so restart Chrome when the update prompt appears.
• Assume your applications are exploitable and test your own software before someone else does; no one finds every vulnerability.
• Plan for the attacks that get through anyway, rather than relying on prevention alone.
• Watch vendor advisories and security news for confirmation and for any further detail, since the technical write-up here is XBOW's own account of its research.

Sources

• XBOW — the firm's own write-up describing the Chrome full-chain exploit, the use of a public OpenAI model, the bounty, and its recommended mitigations.
• The Hacker News — reports that researchers used a public OpenAI model to build the exploit.