As reported by BleepingComputer, a critical use-after-free vulnerability in the Linux kernel's IPv4 TCP networking stack — CVE-2026-98432 — has been demonstrated as remotely exploitable for root-level code execution on servers running Multipath TCP (MPTCP) listeners. The most alarming aspect isn't just the vulnerability itself, but that a working KASLR-bypassing exploit was developed by an AI model in approximately 53 minutes.

Security Impact: As reported by BleepingComputer, a critical use-after-free vulnerability in the Linux kernel's IPv4 TCP networking stack — CVE-2026-98432 — has been demonstrated as remotely exploitable for root-level code execution on servers running Multipath TCP (MPTCP) listeners.

Vulnerability Overview

CVECVE-2026-98432
CVSS (Linux Kernel Team)9.8 / Critical
CVSS (Red Hat)5.9 / Medium (citing exploitation complexity)
TypeUse-after-free via race condition in TCP connection handling
Affected VersionsLinux kernel 4.4 and later (Ubuntu, RHEL, Rocky Linux confirmed)
Confirmed RCE VectorServers with MPTCP listeners enabled
Additional Trigger Conditionsnet.ipv4.tcp_migrate_req enabled or BPF_SK_REUSEPORT_SELECT_OR_MIGRATE programs in use
Active ExploitationNot yet observed in the wild (proof-of-concept exists)
Patch StatusRefer to vendor advisories — check distribution-specific kernel updates

Why This Matters Beyond the CVSS Disagreement

The divergent severity ratings between the Linux kernel CVE team (9.8) and Red Hat (5.9) highlight a persistent industry tension: theoretical exploitability versus practical likelihood. Red Hat's lower score reflects the genuine complexity of triggering the race condition reliably. However, Shield53's assessment aligns closer to the kernel team's rating for one critical reason — the exploitation barrier has been demonstrably collapsed by AI automation.

Anthropic's demonstration that an AI model could develop a working, KASLR-defeating kernel exploit in under an hour fundamentally shifts the economics of offensive security. Vulnerabilities once dismissed as "theoretically exploitable but impractical" must now be reassessed. When AI compresses weeks of expert exploit development into minutes, the gap between vulnerability disclosure and weaponized exploitation narrows dramatically.

The threat model has changed: we are no longer estimating how long a skilled human takes to develop an exploit — we must estimate how long an AI agent takes.

Who Is Most Exposed

Who Is Most Exposed
MPTCP-enabled servers: Any deployment actively using Multipath TCP listeners — increasingly common in telecommunications, CDN edge nodes, and multi-homed data center infrastructure
Load balancers using socket migration: Systems with tcp_migrate_req enabled or BPF reuseport migration programs fall within the additional trigger path
Long-running servers with stable kernels (4.4+): Legacy and LTS deployments that haven't received recent kernel updates are especially vulnerable, as this flaw has existed since kernel 4.4
Internet-facing TCP services: The vulnerability requires no authentication — any reachable TCP listener on an affected kernel is a potential target for denial-of-service; MPTCP listeners face full RCE risk

Shield53 Recommendations

Immediate Actions

  • Patch immediately: Apply the latest kernel update from your distribution vendor. Prioritize any server running MPTCP listeners or socket migration features.
  • Disable MPTCP where not required: If MPTCP is not operationally necessary, disable it via sysctl net.mptcp.enabled=0 to eliminate the confirmed RCE attack surface.
  • Disable TCP request socket migration: Set net.ipv4.tcp_migrate_req=0 unless explicitly required by your load-balancing architecture.
  • Audit BPF reuseport programs: Identify and assess any BPF_SK_REUSEPORT_SELECT_OR_MIGRATE programs in production. These represent an alternate trigger path.
  • Network-level mitigation: Restrict inbound TCP access to internet-facing services using firewall rules and VPN/tunnel overlays where possible. The vulnerability is unauthenticated — network exposure is the primary controllable variable.

Strategic Actions

  • Reassess "complex-to-exploit" advisories: Vulnerabilities previously deprioritized due to exploitation difficulty warrant re-evaluation. AI-assisted exploit development erodes the protection that complexity once provided.
  • Accelerate kernel patch SLAs: The window between disclosure and weaponization is shrinking. Organizations still operating on 30-90 day patch cycles for kernel updates should treat this as an existential risk and move to 7-day or shorter cycles for critical network-stack vulnerabilities.
  • Monitor for exploit publication: While no in-the-wild exploitation has been observed, the Anthropic proof-of-concept demonstrates feasibility. Threat intelligence monitoring for weaponized variants should be elevated.

The Broader Signal: AI as Offense Accelerant

The most significant takeaway from this disclosure is not the kernel bug itself — use-after-free conditions in the TCP stack will continue to be found. The real story is the proof that AI can now produce reliable kernel-level exploits with KASLR bypasses in under an hour. This is a force multiplier for threat actors who previously needed elite exploit developers. Organizations must assume that the time between public vulnerability disclosure and functional exploit availability will continue to shrink, and adjust their defensive posture accordingly: faster patching, reduced attack surface, and defense-in-depth architectures that assume the perimeter will be breached.