Security researcher **Malcolm Stagg** has unveiled a significant new attack class named **NatJack**, which exploits vulnerabilities in Network Address Translation (NAT) connection states. The attack, presented at **Black Hat USA 2026**, demonstrates how an attacker can manipulate NAT connection state to hijack active TCP sessions, spoof DNS responses, expose mapped ports, and even exhaust NAT tables. The research indicates that this behavior affects independently developed NAT implementations across various platforms, including **Windows** and **Linux**. ### CVEs Assigned for Specific Implementations Two implementation-specific flaws have been assigned CVEs: * **CVE-2026-56181** (CVSS score: 8.3): Affects **Windows NAT** as used by **Hyper-V**. * **CVE-2026-63913** (CVSS score: 8.2): Impacts **Linux Netfilter conntrack**. ### Attack Prerequisites and Mitigation Strategies **NatJack** generally requires the attacker to have privileged access to a system operating behind the same NAT as the victim. Consequently, mitigation guidance emphasizes the crucial need to separate untrusted workloads from trusted systems that share NAT infrastructure. There is no single, overarching patch for the entire attack class. Organizations are advised to apply all available **Windows** and **Linux** updates, and critically, to encrypt traffic even within internal networks. The research also recommends the implementation of **Internet Protocol (IP) Source Guard** where applicable. ### How NatJack Exploits NAT Assumptions  The **NatJack** research, conducted independently by Stagg through **SODIUM-24**, targets a fundamental assumption in many NAT implementations: that hosts behind the same NAT will not manipulate each other's connection state. An attacker controlling a system within the same NAT environment can, depending on the specific implementation, manipulate connection-tracking entries belonging to another system. The research outlines four primary attack vectors: 1. **TCP Connection Redirection**: Traffic from an active TCP connection can be redirected by replacing its NAT mapping. 2. **DNS Response Interference**: An attacker can interfere with a victim's DNS request, causing the legitimate DNS response to be routed to the attacker, who then sends a forged response back to the victim.  3. **Port Disclosure**: Techniques can be used to disclose externally mapped ports. 4. **NAT Table Exhaustion**: Attackers can flood the NAT connection table with spoofed flows, preventing legitimate clients from establishing new connections. **Synack** reported that Stagg successfully tested these techniques against numerous real-world network infrastructure products from multiple vendors and demonstrated proof-of-concept exploitation in controlled environments. As of August 7, 2026, there is no public evidence of **NatJack** techniques being exploited in the wild. ### Technical Details of CVEs For **Linux**, the **kernel.org CNA record** for **CVE-2026-63913** details that a crafted SYN packet followed by a reset packet with an invalid sequence number can prematurely force an active **Netfilter NAT** entry into a closed state. This occurs because the conntrack logic failed to validate its direction. Fixed stable releases include 5.10.259, 5.15.210, 6.1.176, 6.6.143, 6.12.93, 6.18.35, 7.0.12, and 7.1. Stagg notes that while the kernel change addresses the code flaw, it only mitigates the broader downstream-spoofing technique, increasing attack complexity rather than fully resolving it. **Microsoft's CNA record** for **CVE-2026-56181** describes the **Windows** issue as an origin-validation error enabling spoofing from an adjacent network. Affected releases include **Windows 11 24H2** before 26100.8875, **25H2** before 26200.8875, **26H1** before 28000.2525, and **Windows Server 2025** before 26100.33158. ### Building on Previous Research **NatJack** builds upon earlier research into NAT-state manipulation, including an **NDSS 2024 study** that demonstrated TCP hijacking through NAT mapping manipulation. That study found 52 of 67 tested routers susceptible to its attack, leading to the assignment of ten CVEs. This ongoing research underscores the persistent vulnerabilities in fundamental network infrastructure components.