Security Firm XBOW Crafts Six-Day Chrome Exploit Using Public AI Model

A team of researchers at cybersecurity company XBOW announced they have assembled a full‑chain exploit for Google Chrome in just six days, leveraging a publicly available OpenAI model to automate the discovery and integration of three previously unknown bugs.

The exploit chains three distinct vulnerabilities in Chrome's rendering and sandbox layers, allowing malicious code to execute on a Windows machine the moment a victim loads a crafted web page. Because the attack requires only a standard page load, it could be delivered via phishing emails, compromised advertisements, or any site that manages to lure a user into visiting the malicious URL.

XBOW’s engineers described the process as a “human‑in‑the‑loop” workflow. They prompted the AI to generate code snippets that probe Chrome’s internals, then used the model’s output to guide automated fuzzing tools toward promising code paths. When the AI suggested promising payloads, the researchers manually verified and refined them, eventually stitching the three bugs together into a single, functional chain.

Chrome’s security architecture relies on multiple isolation mechanisms—process sandboxing, site isolation, and strict memory safety checks—to limit the impact of any single flaw. By linking three separate bugs, the XBOW team was able to bypass these layers, achieving remote code execution without any user interaction beyond loading a page. The researchers did not disclose the specific CVE identifiers, noting that responsible disclosure is underway with Google.

The rapid development timeline underscores a growing concern in the security community: publicly accessible AI models can accelerate the discovery and exploitation of software vulnerabilities. While AI‑driven assistance has been touted for defensive purposes, XBOW’s demonstration shows that the same tools can be weaponized by attackers with relatively modest expertise.

Google has a long‑standing bug‑bounty program and regularly patches Chrome vulnerabilities, often within days of disclosure. The company’s response to this new chain will likely involve a coordinated release of patches for each of the three underlying bugs, followed by a comprehensive update to the browser’s mitigation stack.

Industry analysts say the incident may prompt browser vendors to rethink how they evaluate and mitigate multi‑vector attack paths. Some experts suggest that integrating AI‑based threat modeling into the development lifecycle could help identify chains of seemingly low‑severity bugs before they become exploitable.

For users, the immediate recommendation remains the same: keep browsers and operating systems up to date, employ reputable security software, and exercise caution when clicking on unfamiliar links. As AI tools become more embedded in both defensive and offensive security workflows, the line between rapid vulnerability research and accelerated weaponization is likely to blur, raising new challenges for the broader cybersecurity ecosystem.