MPTCP

Anthropic's Internal Claude Model Writes Exploit for Linux Kernel CVE-2026-98432 Enabling Remote Root Command Execution

An unreleased internal model from Anthropic generated a working exploit that achieves remote command execution with root privileges on a Linux server configured with multipath TCP (MPTCP). The exploit targets CVE-2026-98432, a use-after-free vulnerability in the Linux kernel network stack that occurs during parallel TCP request processing. Researchers first used the model to bypass KASLR address randomization, a process that took 48 minutes, followed by five additional minutes to seize control and execute arbitrary commands. The Linux kernel team rated the flaw 9.8 on CVSS while Red Hat assigned it 5.9, citing exploitation complexity. A patch has already been merged into the mainline kernel and supported stable branches, though users must verify availability through their distribution vendor. Anthropic emphasized that the experiment relied on a non-public model and cannot be reproduced with standard chat interfaces.