An attacker who already holds code execution on a network can now reach across the NAT boundary and take over neighbors’ connections, redirect their DNS lookups, or simply shut down shared internet access. The attack class, named NatJack and presented at Black Hat USA 2026 by researcher Malcolm Stagg, exploits the long-standing assumption that hosts behind the same NAT will not interfere with one another’s traffic state.
Stagg found the weakness runs deep in how connection tracking tables are built and validated. Because NAT was designed as an address-exhaustion stopgap in the early 1990s, not as a security boundary, many implementations trust peers on the same public IP implicitly. Four attack paths follow from that trust: TCP session hijack, DNS response poisoning, mapped-port disclosure, and table exhaustion that blocks new connections.
Every one of the 32 products and configurations Stagg tested across 13 vendors was vulnerable to at least one technique, and the problem spans Windows, Linux, and macOS, which share no common NAT codebase. Two specific implementations got CVEs: CVE-2026-56181 (CVSS 8.3) in Windows NAT for Hyper-V, and CVE-2026-63913 (CVSS 8.2) in Linux Netfilter conntrack, whose fix landed in stable kernels from 5.10.259 through 7.1.
The bar for exploitation is a foothold on a system behind the same NAT as the target, which makes shared cloud infrastructure and multi-tenant gateways the highest-risk settings. As of August 7, no public evidence tied NatJack to real-world intrusions.
Vendors disagree on whether this is a bug at all, with Cisco and Apple classifying the behavior as a design limitation rather than a vulnerability. Encryption blunts but does not remove the risk. Practical mitigations include separating untrusted workloads onto their own NAT gateways, enabling IP Source Guard, encrypting internal traffic, and watching for saturated NAT tables, port floods, and duplicated source IPs.
