CVE-2026-123633: Critical NGINX HTTP/2 Use-After-Free Flaw Enables RCE

CVE-2026-123633: Critical NGINX HTTP/2 Use-After-Free Flaw Enables RCE

SOC Prime Team
SOC Prime Team linkedin icon Follow

A newly disclosed vulnerability tracked as CVE-2026-123633 exposes certain NGINX reverse-proxy deployments to unauthenticated denial-of-service attacks and potentially remote code execution. The flaw is a use-after-free condition in NGINX’s HTTP/2 processing that can be triggered when the server proxies client requests to an upstream backend while a worker is running low on available connections.

The vulnerability was disclosed alongside CVE-2026-123007 as part of a broader set of security flaws patched by the NGINX team on November 18, 2026. F5 assigns CVE-2026-123633 a CVSS v4.0 score of 9.2, Critical, while its CVSS v3.1 score is 8.1, High because exploitation requires a particular server state and configuration.

Unlike CVE-2026-123007, which can affect HTTP/2-enabled NGINX instances serving ordinary static content, CVE-2026-123633 specifically targets reverse-proxy processing. An exposed server must have HTTP/2 enabled, forward requests to a backend server, and reach a state where available worker_connections are running low.

Successful exploitation can crash NGINX worker processes or corrupt memory. Wiz researchers demonstrated that the vulnerability can be developed into unauthenticated remote code execution and confirmed an ASLR bypass in laboratory environments. Code execution occurs with the privileges of the NGINX worker process rather than automatically granting root access to the host.

Organizations can use SOC Prime’s AI-Native Detection Intelligence Platform to operationalize vulnerability intelligence and accelerate threat hunting and detection engineering across SIEM, EDR, XDR, and Data Lake environments.

Contact Sales

Detect CVE-2026-123633 by correlating abnormal HTTP/2 traffic and connection exhaustion with NGINX worker crashes, suspicious process execution, and unexpected network connections originating from NGINX processes. Because exploitation occurs in the web-server process itself, combining network and endpoint telemetry provides substantially stronger visibility than relying on static indicators alone.

CVE-2026-123633 analysis

CVE-2026-123633 affects NGINX Open Source versions beginning with 1.19.7 through 1.31.8 when HTTP/2 is enabled and requests are proxied to an upstream backend. NGINX Open Source 1.30.7 and 1.31.9 contain the fix. NGINX Plus is also affected, with fixes for this specific vulnerability introduced in 37.0.7.1 and 37.2.0.1.

The underlying weakness is a use-after-free condition in NGINX’s HTTP/2 connection management. NGINX keeps a list of connections marked as reusable so it can close them when additional connection slots are required. A connection should only remain on that list when it has no active request being processed.

Researchers found that an attacker can initially send only part of an HTTP/2 HEADERS frame. Because no active stream has yet been created, NGINX marks the connection as reusable while waiting for the remainder of the frame.

After the attacker sends the remaining portion, NGINX creates an HTTP/2 request stream and begins processing the request. However, the connection can incorrectly remain marked as reusable even though it is now actively handling a request. This creates the precondition for the memory-safety flaw.

The attacker then causes the request to be forwarded to an upstream application server. Establishing that backend connection requires NGINX to obtain another connection slot. When the worker approaches its configured worker_connections limit, NGINX can invoke its connection-draining logic to free capacity.

Because the malicious HTTP/2 connection was incorrectly left on the reusable list, NGINX can select and close that same connection even while its request is still being processed. During connection cleanup, memory associated with the HTTP/2 connection can be freed while other portions of the request-processing path still retain references to it.

Execution subsequently continues through those stale pointers, producing a classic use-after-free condition. If the freed memory is reallocated for attacker-influenced data, the NGINX worker can interpret new contents as the original HTTP/2 request or connection structures.

This memory corruption can initially cause a segmentation fault or worker crash. More advanced exploitation requires manipulating memory allocation so that attacker-controlled data occupies the freed region. Researchers showed that the primitive can be extended to disclose memory addresses, manipulate heap state, defeat ASLR, and ultimately achieve arbitrary code execution.

F5’s vulnerability description states that code execution is possible when ASLR is disabled or when an attacker can bypass it. Wiz researchers went further and reported successfully developing an ASLR bypass, demonstrating that ASLR should not be considered a sufficient protection against a determined attacker.

The attack is remote and does not require authentication. However, exploitation has several environmental prerequisites: the attacker must communicate directly with an HTTP/2-enabled NGINX listener, that instance must proxy requests to a backend, and the targeted worker must reach a state where free connection slots are scarce.

This makes the flaw particularly relevant to heavily utilized reverse proxies, API gateways, ingress infrastructure, SaaS front ends, and web platforms where NGINX handles large numbers of concurrent HTTP/2 connections while forwarding traffic to application servers.

A CDN or WAF that terminates HTTP/2 before traffic reaches the NGINX origin may disrupt the precise frame behavior required for exploitation. However, researchers caution that protection depends on the intermediary’s implementation and should not be treated as a substitute for updating the underlying NGINX instance.

The CVE-2026-123633 PoC status requires some nuance. Wiz and Xint developed working exploitation techniques and demonstrated remote code execution during coordinated research, but Wiz explicitly stated that it would not release its full exploit. Several GitHub repositories claiming proof-of-concept implementations appeared after disclosure, but their existence should not be interpreted as independent validation of a reliable universal RCE exploit.

Xint reports that its researchers discovered the group of NGINX vulnerabilities on October 11, 2026 and submitted them to F5 through HackerOne on October 16. Wiz independently reported the HTTP/2 use-after-free later assigned this CVE on October 21, including confirmed RCE and a proposed fix. F5 provided a patch for validation on November 10, followed by coordinated public disclosure on November 18.

The vulnerability had been present since NGINX 1.19.7, meaning the affected code path remained in production releases for nearly six years before disclosure.

For CVE-2026-123633 detection, security teams should monitor for abnormal worker exits, segmentation faults, sustained HTTP/2 connection pressure, and unusual conditions around connection exhaustion. A log entry indicating an NGINX worker exited on a signal can provide evidence of a failed exploitation attempt, although worker crashes alone are not high-confidence proof of malicious activity.

Endpoint telemetry becomes especially important for detecting successful exploitation. NGINX worker processes unexpectedly launching sh, bash, curl, wget, Python, Perl, or other command interpreters should be treated as highly suspicious. Defenders should also investigate new outbound network connections, file downloads, persistence activity, or credential access initiated by NGINX worker processes.

Network teams should monitor scenarios where one or more sources maintain large numbers of HTTP/2 connections while the affected server simultaneously experiences connection pressure or unexpected worker failures. Because the use-after-free depends on connection-draining behavior, this correlation may provide greater fidelity than simply alerting on high HTTP/2 volume.

Traditional CVE-2026-123633 IOCs such as malware hashes, fixed attacker domains, or known command-and-control servers are currently of limited value. The vulnerability can be exploited directly over HTTP/2 without requiring a predefined malware payload or attacker-controlled infrastructure, making behavior-based monitoring and exposure management more effective.

As of November 19, 2026, F5 and the reviewed research teams have not reported confirmed exploitation in real-world attacks, and the vulnerability is not listed in CISA’s Known Exploited Vulnerabilities catalog. Nevertheless, working private exploits, detailed technical research, and public interest in the flaw make rapid remediation advisable.

Security teams seeking further details for CVE-2026-123633 should review their exact NGINX topology instead of relying only on version scans. Confirm whether HTTP/2 is accepted on the origin, whether NGINX proxies those requests to upstream servers, and whether a CDN or another proxy terminates HTTP/2 before the traffic reaches NGINX.

CVE-2026-123633 Mitigation

The primary CVE-2026-123633 mitigation is to upgrade NGINX immediately to a release containing the security correction.

ProductVulnerable versionsFixed version
NGINX Open Source1.19.7–1.30.61.30.7
NGINX Open Source mainline1.31.0–1.31.81.31.9
NGINX Plus 37.0.xBefore 37.0.7.137.0.7.1
NGINX Plus 37.1/37.2Before 37.2.0.137.2.0.1

NGINX Open Source users should preferably upgrade to 1.30.7 or 1.31.9. Organizations running NGINX Plus should note that the versions fixing this particular use-after-free are older than the releases required to fix CVE-2026-123007 as well. Moving to the latest vendor-recommended security build therefore provides broader protection against the vulnerabilities disclosed in the November security release.

After installing an updated package, administrators should ensure the new binary is actually running. The Hacker News notes that a normal configuration reload may leave existing workers running the previous executable. Perform a full restart or appropriate live binary upgrade and confirm that old worker processes have exited.

If immediate patching is impossible, disabling HTTP/2 blocks the known attack path for this issue. On newer NGINX releases, administrators can configure:

http2 off;

Legacy configurations should also remove the http2 parameter from affected listen directives. Red Hat recommends this workaround for exposed deployments that cannot yet be updated.

Organizations should test application compatibility before disabling HTTP/2 because native gRPC and other services can depend on HTTP/2 functionality. The workaround should therefore be treated as temporary rather than a permanent replacement for the security update.

Administrators should inventory NGINX deployments acting as reverse proxies, ingress controllers, API gateways, or application front ends. Prioritize internet-facing instances that expose HTTP/2 and forward traffic to upstream application servers, since this combination provides the primary vulnerable configuration.

Teams should also examine the worker_connections configuration and historical utilization. Raising the limit may make the specific exploitation condition more difficult to trigger, but it does not correct the underlying use-after-free and should not be treated as a security fix.

Following remediation, review NGINX error logs and endpoint telemetry from the period when the system was exposed. Look for unexplained worker crashes, processes launched by NGINX, suspicious file creation, or unexpected outbound connections. A server that shows evidence of successful process execution should be handled as potentially compromised rather than considered safe merely because it has since been patched.

FAQ

What is CVE-2026-123633 and how does it work?

CVE-2026-123633 is a use-after-free vulnerability in NGINX’s HTTP/2 processing. An attacker can manipulate an HTTP/2 connection so NGINX incorrectly keeps it marked as reusable while a request is being processed. When the server is short on connection slots and attempts to establish an upstream connection, connection-draining logic can close and free the active HTTP/2 connection. Subsequent request processing then accesses the freed memory, potentially causing a crash, memory corruption, or remote code execution.

When was CVE-2026-123633 first discovered?

Xint reports discovering the wider set of related NGINX flaws on October 11, 2026 and reporting them to F5 on October 16. Wiz independently identified the HTTP/2 use-after-free and submitted it to F5 on October 21 with a working RCE demonstration and proposed patch. The vulnerability was publicly disclosed on November 18, 2026 alongside the NGINX security releases.

What is the impact of CVE-2026-123633 on systems?

Successful exploitation can crash NGINX workers and cause denial of service. Researchers also demonstrated memory disclosure, heap corruption, ASLR bypass, and unauthenticated arbitrary code execution with the privileges of the NGINX worker. Because NGINX often fronts multiple websites, APIs, and applications, compromising one reverse proxy can potentially expose several downstream services.

Can CVE-2026-123633 still affect me in 2026?

Yes. A deployment can remain vulnerable if it runs an affected NGINX version, accepts HTTP/2 connections, proxies those requests to a backend server, and has not received the vendor fix. NGINX Open Source releases from 1.19.7 through 1.31.8 contain the vulnerable code. Organizations using vendor-packaged or embedded NGINX builds should verify their vendor’s backport status rather than relying solely on the upstream version string.

How can I protect myself from CVE-2026-123633?

Upgrade NGINX Open Source to 1.30.7 or 1.31.9 and install the appropriate fixed NGINX Plus release. Restart the service so vulnerable worker processes are terminated. If an immediate update is impossible, temporarily disable HTTP/2 on exposed listeners and verify that clients can no longer negotiate h2 with the NGINX origin. Finally, review historical logs and endpoint telemetry for worker crashes or unexpected command execution that could indicate exploitation attempts.

Join SOC Prime's Detection as Code platform to improve visibility into threats most relevant to your business. To help you get started and drive immediate value, book a meeting now with SOC Prime experts.

More CVEs Articles