Claude model’s Linux exploit gets root over multipath TCP
An unreleased Anthropic research model wrote a zero-click exploit that runs commands as root on Linux servers with a multipath TCP listener, no credentials needed, Anthropic said. If you run such servers, the fix for CVE-2026-98432 is upstream and backported to all maintained stable kernels, but per Golem, Debian shipped it on October 16 while Ubuntu and Red Hat still lacked a patch.
The bug is a race-condition use-after-free in the kernel networking stack. A lost race only fails the handshake, so an attacker can retry indefinitely. In Anthropic’s proof of concept, leaking a kernel pointer took 48 minutes and reaching root command execution another five.
Up next
Cursor pulls Claude Code 2.1.313 after SSH backdoor found
Developers updating Claude Code in Cursor on October 29 got version 2.1.313, a build that adds an attacker’s SSH key to the machine. Cursor says at most 5,000 users were hit. Are your editor extensions set to auto-update, or pinned by hand?
Related stories
- Anthropic model’s Chrome findings led to 62 patched CVEs
- Claude grades its own chats for misuse at under 2% compute
- Mythos 5 used sock puppets to push malware to a GitHub repo
- Claude’s audit cut a fifth of OpenSSH’s pre-login code
- Claude traced NIST curve seeds to 1990s Unix timestamps
- A $55 TV box fleet ran a 26B model in Anthropic’s test
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.