Security researcher Malcolm Stagg has revealed a new class of attack called NatJack that exploits network address translation systems to take over active TCP sessions, forge DNS responses, reveal mapped ports, and overwhelm NAT tables. Presented at Black Hat USA 2026, the research uncovered vulnerable behavior across multiple independently built implementations, spanning both Windows and Linux platforms. Two specific flaws received CVE identifiers: CVE-2026-56181 with a CVSS score of 8.3 in Windows NAT used by Hyper-V, and CVE-2026-63913 with a CVSS score of 8.2 in Linux Netfilter conntrack.
The attack generally requires an attacker to hold privileged access to a machine located behind the same NAT as the target. Stagg tested the techniques against dozens of real-world network infrastructure products from various vendors and showed proof-of-concept exploitation in a controlled setting, according to Synack. For Linux systems, the kernel.org CNA record indicates that a specially crafted SYN packet followed by a reset with an invalid sequence number can force an active Netfilter NAT entry prematurely into a closed state because the conntrack logic didn't validate its direction. Fixed stable releases span versions 5.10.259, 5.15.210, 6.1.176, 6.6.143, 6.12.93, 6.18.35, 7.0.12, and 7.1. Microsoft's CNA record characterizes the Windows problem as an origin-validation error that permits spoofing from an adjacent network, affecting Windows 11 24H2 before build 26100.8875, 25H2 before 26200.8875, 26H1 before 28000.2525, and Windows Server 2025 before 26100.33158.
The research describes four main attack paths, according to the report. One method redirects traffic from an active TCP connection by replacing its NAT mapping. Another interferes with a victim's DNS request so the legitimate DNS response reaches the attacker, enabling transmission of a forged response. Additional techniques disclose externally mapped ports or fill the NAT connection table with spoofed flows until legitimate clients can't establish new connections. Stagg says the kernel change fixes the code flaw but only mitigates the broader downstream-spoofing technique, raising attack complexity rather than fully addressing it. The Hacker News found no public evidence that NatJack techniques have been exploited in the wild as of August 7, 2026.
The NatJack research, conducted independently by Stagg through SODIUM-24, targets an assumption embedded in many NAT implementations: hosts behind the same NAT are generally assumed not to manipulate one another's connection state. An attacker controlling a system behind the same NAT can, depending on the implementation, manipulate connection-tracking entries belonging to another system. The research also builds on earlier work into NAT-state manipulation, including an NDSS 2024 study that showed TCP hijacking through NAT mapping manipulation and found 52 of 67 tested routers susceptible to its attack, producing ten CVEs. There's no single patch for the broader attack class. Organizations should deploy available Windows and Linux updates and encrypt traffic even within internal networks, the mitigation guidance emphasizes. The research also recommends Internet Protocol Source Guard where applicable, and the guidance stresses separating untrusted workloads from trusted systems that share NAT infrastructure. The underlying architectural assumption that co-located hosts won't attack each other means enterprises must rethink internal network trust boundaries whenever virtualization or multi-tenant infrastructure places diverse workloads behind shared translation layers. Organizations accustomed to treating internal zones as inherently safe may find that model no longer holds once an adversary gains even limited initial access.