Claude Code Uncovers Linux Kernel Vulnerability Hidden for 23 Years

Available in: 中文
2026-04-04T11:31:40.387Z·1 min read
Anthropic research scientist Nicholas Carlini revealed at the unprompted AI security conference that he used Claude Code to discover multiple remotely exploitable security vulnerabilities in the Li...

AI-Powered Security Research Yields Historic Discovery

Anthropic research scientist Nicholas Carlini revealed at the unprompted AI security conference that he used Claude Code to discover multiple remotely exploitable security vulnerabilities in the Linux kernel including one that remained undiscovered for 23 years.

The Breakthrough

Carlini expressed astonishment at Claude Code effectiveness stating he had never found a remotely exploitable heap buffer overflow in the Linux kernel before. With the AI coding assistant he found multiple such vulnerabilities which he described as extremely difficult to discover through traditional methods.

Methodology

The approach was remarkably simple. Carlini wrote a script that iterates over every file in the Linux kernel source tree and instructs Claude Code to treat each file as a CTF challenge. The AI was asked to find vulnerabilities and write the most serious one to a report file.

This brute-force approach combined with Claude Code deep understanding of complex systems proved devastatingly effective.

The NFS Vulnerability

The most notable discovery was a bug in the network file system NFS driver that allows an attacker to read sensitive kernel memory over the network. This vulnerability required understanding intricate NFS protocol details including:

The attack requires two cooperating NFS clients working together to trigger the memory read.

Implications for Cybersecurity

This discovery demonstrates that AI coding assistants can fundamentally transform security auditing. What previously required years of specialized expertise can now be partially automated allowing AI to systematically review millions of lines of code for vulnerabilities that human auditors may have missed for decades.

Source: mtlynch.io https://mtlynch.io/claude-code-found-linux-vulnerability/

← Previous: Anthropic Discovers Functional Emotion Representations in Claude Sonnet 4.5Next: US March Nonfarm Payrolls Surge to 178,000, Unemployment Drops to 4.3% →
Comments0