Claude-Built Linux Exploit Targets Bug That Ubuntu Still Hasn't Patched
Summary
Anthropic says its unreleased AI model generated a working exploit for a Linux kernel flaw, CVE-2026-98432, rated 9.8 in severity. The attack can potentially give a remote attacker root access without a password, exploiting a race condition; the demonstrated method requires a server with an MPTCP listener and can be retried with little risk of crashing the system. The vulnerability affects Linux versions dating back to 4.4. Red Hat and several other distributions have released fixes for some versions, while Ubuntu lists multiple releases as vulnerable, with fixes in progress for newer versions. Red Hat offers a temporary firewall mitigation but says it does not fully prevent exploitation.
