When Tencent Zhuque Lab disclosed CVE-2026-64564 on August 6, the disclosure package included something unusual: attribution to Corvus AI, a multi-agent pipeline built specifically for kernel vulnerability research. The bug itself had been sitting in the Linux kernel's SCTP implementation since approximately 2008. An AI agent found it first.
What SCTPhantom is
SCTPhantom is a use-after-free vulnerability in the Linux kernel's Stream Control Transmission Protocol (SCTP) implementation, specifically in the ASCONF (Address Configuration Change) chunk handling. The root cause is an identity mismatch: when the kernel processes a DEL-IP delete operation, it validates the request using the packet's source address, but a separate cached pointer used to select the network path relies on a different address parameter. This mismatch can be exploited to free kernel memory that is still in use, triggering a use-after-free condition.
The practical consequences are severe. A local attacker can escalate privileges to root. In containerized environments, the flaw enables container-to-host escape under standard configurations, bypassing kernel namespace isolation. Tencent has confirmed successful exploitation on Debian 13, Ubuntu 24.04, RHEL 9, Rocky Linux 9, and OpenCloudOS. The CVSS score is 7.8, with the local privilege escalation vector.
The AI angle: Corvus AI's multi-agent discovery pipeline
Tencent Zhuque Lab attributes the SCTPhantom discovery to Corvus AI, a research pipeline that uses multiple AI agents working in parallel to analyze kernel subsystems for vulnerability patterns. This is at minimum the third major kernel vulnerability surfaced by AI-assisted tooling in 2026, following similar AI-assisted discoveries in the KVM and eBPF subsystems earlier this year.
The pattern across these discoveries is consistent: the bugs are old, sometimes decades-old, sitting in code that has been reviewed many times by human researchers. AI-assisted static analysis appears to find a different class of bug than human audit or fuzzing: specifically, semantic mismatches in reference counting and memory ownership that are difficult to detect through manual code review but are amenable to formal reasoning at scale.
This has a specific implication for defenders: expect an accelerating cadence of long-dormant vulnerability disclosures in foundational Linux kernel subsystems. The pipeline for finding these bugs is now automated. We will see more SCTPhantoms.
Affected systems and patches
The vulnerability affects any Linux kernel with SCTP support enabled (CONFIG_IP_SCTP compiled in or loaded as a module), which is the default configuration on all major distributions. Patches have been backported to stable kernel branches 6.6.148, 6.12.101, 6.18.42, and 7.1.6.
Debian 13 users should apply the latest kernel security update, which addresses both SCTPhantom and the Zapscape KVM escape disclosed the same week. Ubuntu 24.04 LTS received an updated kernel from Canonical. RHEL 9 and Rocky Linux 9 patches are available through standard advisory channels. OpenCloudOS patches were released alongside the disclosure.
What to do
Apply the kernel update for your distribution immediately. If you run containerized workloads, treat this as high priority: the container-to-host escape vector means a compromise at the container level can escalate to the underlying host without additional privilege escalation. For RHEL and Rocky Linux 9 environments that cannot patch immediately, disabling SCTP module loading via modprobe blacklist is an interim mitigation, though it will affect workloads that depend on SCTP.
Check whether SCTP is in use before blacklisting: run lsmod | grep sctp to confirm if the module is loaded. On systems without any SCTP-dependent services, blacklisting is a safe interim measure until the kernel patch can be applied during a maintenance window.
Gigia Tsiklauri is a cybersecurity analyst and the founder of infosec.ge. Reach out to discuss threat intelligence or collaboration.