A Linux kernel flaw nearly two decades old has yielded host root and a container escape in Tencent’s testing, and the researchers credit an AI pipeline for finding it. The bug, named SCTPhantom and cataloged as CVE-2026-64564, has been present in every kernel since 2008.
The vulnerability sits in SCTP’s dynamic address reconfiguration. The kernel validates a delete request against the packet’s source address, then acts on a path selected using a different address inside the message. A message that carries an address, a delete for that address, and a wildcard delete frees the path and reuses the dead pointer, leaving the connection pointing at released memory.
Kernel maintainers shipped the fix August 3 in stable releases 7.1.6, 6.18.42, 6.12.101, and 6.6.148. Tencent rates the vulnerability 8.5 under CVSS v4.0; NVD had not published a severity rating as of August 7.
Exploitation is local and demands that SCTP be reachable on the target, which limits exposure. Where those conditions hold, Zhuque Lab reports root on Debian 13, Ubuntu 24.04, Rocky Linux 9, RHEL 9, and OpenCloudOS. The container escape ran with the default seccomp profile and without CAP_NET_ADMIN or CAP_SYS_ADMIN, reaching host root in six of eight attempts. No outside researcher has reproduced the result, and the lab did not name the container runtime it tested.
For defenders, the practical step is simple: if SCTP is not in use, disable the module and the exposure disappears. Teams that must run it should verify their kernel builds carry the August 3 fix, keeping in mind that the same stable releases also address Zapscape, an unrelated KVM escape disclosed the same day. Tencent says an automated research pipeline called Corvus AI surfaced the bug, the latest in a string of aged kernel flaws that machine-assisted analysis has revived this year.
