Anthropic AI Creates Exploit for Critical Linux System Flaw

Anthropic has revealed that its artificial intelligence model, Claude, created a functional attack against the Linux operating system. The flaw, identified as CVE-2026-98432, allows an attacker to take full control of a server remotely without needing access credentials.

The issue carries a severity score of 9.8 out of 10. This vulnerability has existed since version 4.4 of the Linux kernel and affects a vast range of global servers. The attack exploits a race condition in the system memory that allows for infinite intrusion attempts without causing crashes.

Distributions such as Red Hat, Debian, and Oracle Linux have already released fixes for the error. However, Ubuntu has yet to release a definitive patch for its vulnerable versions. The flaw is particularly dangerous on systems that utilize the multipath TCP functionality.

The exploitation process developed by the AI takes less than an hour to locate the memory flaw. Afterward, the system grants root-level access to the attacker in just a few minutes. The ease of executing machine-generated code increases concerns regarding the security of digital infrastructure.

This advancement in the ability of artificial intelligence to create exploits may force technology companies to accelerate investments in cybersecurity. The automation of complex attacks is likely to raise operational costs for maintaining networks and servers across the market.