Anthropic AI finds 11-year-old Linux kernel vulnerability

Anthropic’s cloud system discovered a critical security vulnerability in the Linux kernel during systematic tests of major projects. The flaw, which allowed for remote code execution, had remained undetected in the codebase for exactly 11 years, affecting millions of systems. Linux distributors are now urgently working to provide updated kernels for affected supported products.

Jürgen Schmidt of Heise Security described the discovery as a striking example of how AI is uncovering security flaws in decades-old codebases. He noted that while Linux network code is carefully maintained, most software is less robust, and previous protection relied on the difficulty of finding and exploiting such holes. Schmidt warned that large language models are now systematically and largely automatically exposing these vulnerabilities. He indicated that a massive wave of weaknesses is approaching, signaling a significant shift in cybersecurity as AI tools make it easier to identify flaws that humans previously overlooked.