April 4, 2026
AI finds a 23‑year bug—devs lose it
Claude Code Found a Linux Vulnerability Hidden for 23 Years
AI sidekick digs up a 23‑year Linux bug — fans cheer, skeptics sneer
TLDR: An Anthropic researcher used Claude Code to uncover multiple Linux kernel flaws, including a 23‑year‑old bug in file sharing, with minimal hand‑holding. The community split fast: jokes about “million‑token eyeballs,” worries about AI costs and new risks, and a debate over whether this is genius or just finally looking closely.
Anthropic researcher Nicholas Carlini says his coding copilot, Claude Code, poked through the Linux kernel and surfaced multiple remotely exploitable flaws — including a 23‑year‑old ghost hiding in the network file sharing system (NFS). The twist the internet loves: he mostly just pointed the AI at the code and asked, “Where are the bugs?” Cue the drama.
The hype squad is loud: commenters are remixing Linus’s famous line into “given a million‑token context window, all bugs are shallow,” with dist‑epoch earning upvotes for the meme. Others are clutching their wallets. Jazz9k rains on the parade with “the cost of tokens will prevent most companies” from using AI watchdogs 24/7. Then the cynics roll in: up2isomorphism warns Claude could “introduce more vulnerability than it discovered,” while userbinator nitpicks the headline — not “hidden,” just “no one bothered to look.” Translation: was this genius sleuthing or finally shining a flashlight under the couch?
Meanwhile, the technical flex that got everyone’s attention: the bug wasn’t an obvious typo — it involved NFS’s nitty‑gritty and let attackers peek at sensitive memory. The vibe? AI bounty hunter vs. human pride. Some call it a watershed moment for code security; others say don’t crown the robot yet. Also trending: NFS now stands for “New Found Secrets.”
Key Points
- •Anthropic researcher Nicholas Carlini used Claude Code to find multiple remotely exploitable Linux kernel vulnerabilities, including one present for 23 years.
- •He presented the work at an AI security conference, noting the difficulty of finding such bugs without language model assistance.
- •Carlini used a simple script to iterate over every kernel source file, prompting Claude Code (framed as a CTF) to locate serious vulnerabilities and write reports.
- •To avoid duplicates, the script focused the model on different files per iteration, using a command-line workflow with verbose and special flags.
- •A highlighted vulnerability in the Linux NFS driver could let attackers read sensitive kernel memory; the example involved two clients and NFS operations with a 1024-byte owner ID.