Anthropic’s Claude writes Linux exploit for unpatched CVE-2026-98432
Anthropic disclosed on October 27 that its AI model generated a working exploit for CVE-2026-98432, a critical Linux vulnerability scoring 9.8/10 in severity. The bug, present since Linux 4.4, lets attackers gain full server control without passwords by exploiting a race condition in memory handling. Anthropic’s exploit uses an unreleased internal Claude model and targets systems with multipath…
Key points
- Anthropic’s AI wrote an exploit for **CVE-2026-98432**, a Linux vulnerability scoring **9.8/10** in severity
- Attackers can gain root access in under **1 hour** by exploiting a memory race condition in **multipath TCP**
- Ubuntu still lists **16.04–26.10** as vulnerable, with fixes only confirmed for **22.04**, **24.04**, **26.04**, and **26.10**
Ubuntu’s tracker still marks versions 16.04 through 26.10 as vulnerable, with fixes only confirmed for 22.04, 24.04, 26.04, and 26.10. Red Hat, Debian, SUSE, Oracle Linux, and Amazon Linux have released patches for some versions, but Ubuntu has not yet issued fixes for 20.04, 18.04, or 16.04. A temporary workaround from Red Hat reduces exposure but does not fully block attacks. Anthropic previously used its Claude Mythos model to help Mozilla discover 271 bugs in Firefox.
The story so far
2 episodes →- Anthropic’s Claude writes Linux exploit for unpatched CVE-2026-98432this story
Claude-Built Linux Exploit Targets Bug That Ubuntu Still Hasn't Patched
tech.yahoo.com · 2 November 2026
Loading the full article…
This text was published by tech.yahoo.com and written by Jose Antonio Lanz. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Society & Work
All →- Reddit users share how to spot AI-generated voice clones with free tools · 1 src
- Microsoft Copilot speculates Pudgy Penguins may partner with a major brand · 1 src
- Proton users cancel subscriptions after CEO dismisses security testing as 'loser thinking' · 3 src
- Hackers impersonate AI bots to hunt for credentials in credential-hunting campaign · 1 src
- Microsoft patents AI to edit live speech and faces in real time · 4 src
Comments
via GitHub Discussions