NatJack Attack Exploits NAT Flaws to Hijack TCP Sessions and Manipulate DNS Traffic

Security researchers have disclosed a new network attack class called NatJack, revealing fundamental weaknesses in Network Address Translation (NAT) state management that could allow attackers to hijack TCP sessions, manipulate DNS responses, disclose externally mapped ports, and disrupt network connectivity.

Presented by security researcher Malcolm Stagg at Black Hat USA 2026, the research suggests the problem extends beyond a single vendor or operating system because NatJack targets long-standing assumptions in how NAT infrastructure tracks connections.

NAT is widely deployed across enterprise networks, home routers, cloud environments, containers, and virtualization platforms. A NAT device maintains connection-tracking tables that map internal private IP addresses and ports to external addresses and ports.

NatJack Attack Exploits NAT Flaws

These mappings allow multiple systems to communicate through shared public infrastructure. NatJack challenges the assumption that systems located behind the same NAT boundary cannot maliciously interfere with connection state belonging to other hosts.

Testing reportedly identified vulnerable behavior across 32 products and configurations, including routers, firewalls, cloud platforms, container environments, and hypervisors.

The attack manipulates NAT connection-tracking behavior by sending carefully constructed network packets capable of removing, replacing, or interfering with mappings associated with another system.

One of NatJack’s most significant capabilities involves TCP session hijacking through upstream or downstream packet spoofing.

In one demonstrated scenario, an attacker sends crafted traffic that causes a NAT implementation to interpret a victim’s existing connection as terminated.

Once the legitimate mapping disappears, the attacker attempts to establish a replacement mapping associated with the victim’s external address and port.

Depending on the affected implementation and application protocol, successful exploitation could allow an attacker to inject traffic, disrupt communications, or potentially intercept portions of an established session.

Researchers highlighted long-lived HTTP connections as one potential target. Applications protected by properly validated TLS would provide an additional security boundary against meaningful traffic manipulation.

NatJack is particularly notable because it does not necessarily depend on traditional Layer 2 attacks such as ARP spoofing.

An attacker and victim could potentially exist on separate VLANs, subnets, or broadcast domains while remaining vulnerable if their traffic ultimately traverses shared NAT infrastructure.

Consequently, conventional switch-level isolation alone might not adequately protect multi-tenant environments against this attack class.

Researchers also demonstrated UDP DNS response manipulation. By interfering with connection-tracking state associated with a victim’s outstanding DNS query, an attacker could potentially redirect the legitimate response and inject a forged DNS response.

Where authenticated DNS protections are absent, successful exploitation could redirect users or applications toward attacker-controlled infrastructure.

Another NatJack technique focuses on NAT table exhaustion. Attackers can generate large numbers of spoofed flows designed to consume available connection-tracking entries.

Once the NAT state table reaches capacity, legitimate clients may be unable to establish new connections, resulting in a denial-of-service condition.

Researchers additionally documented techniques for discovering externally mapped ports, potentially providing information useful for more targeted attacks.

The security implications are especially relevant to environments where trusted and untrusted workloads share NAT infrastructure, including corporate networks, multi-tenant cloud environments, Docker hosts, Kubernetes nodes, hypervisors, virtual machines, and serverless platforms.

Home networks could also face exposure when an attacker already controls a malicious or compromised device behind the same router.

Two implementation-specific vulnerabilities associated with the research have reportedly received identifiers: CVE-2026-56181, affecting Windows NAT and certain Hyper-V configurations, and CVE-2026-63913, involving Linux Netfilter connection tracking.

However, implementation patches may address individual exploitation techniques without eliminating the broader architectural attack class.

Organizations should prioritize relevant vendor security updates while reviewing how untrusted systems share NAT infrastructure. Defensive measures should include workload isolation, anti-spoofing controls such as IP Source Guard, NAT-state monitoring, and detection of anomalous SYN/RST traffic.

Encrypting application communications with TLS and deploying authenticated DNS protections can further limit the consequences of successful manipulation.

NatJack ultimately demonstrates that NAT should not be treated as a security boundary by itself, particularly when potentially hostile workloads share the same translation infrastructure.