Anthropic: Claude identified vulnerability chains in popular systems

Anthropic said its internal research model, Claude, helped discover and chain vulnerabilities in Chrome, Safari, Android, Windows, and Linux. According to the company, some chains made it possible to execute code remotely or access data; the information was shared with developers, and the vulnerabilities were fixed. Anthropic reported the findings on October 27, 2026.

In one example involving Linux, a remote attacker could send packets to a server with multipath TCP enabled and execute commands with root privileges, without credentials or user interaction. In the demonstration, disclosing the kernel address took 48 minutes, and executing the command took another five. The vulnerability, CVE-2026-98432, was fixed in the mainline Linux branch and backported to all supported stable branches.

Translation editors: Соловьев В., Пеков А.