Skip to content

+Threat Wire · Reference

nginx2root: remote code execution in nginx HTTP/2, HTTP/3, resolver and mail proxy

Five memory corruption flaws in nginx, together called nginx2root, can let an unauthenticated attacker crash a worker process or run code on the server, each in a different configuration.

Prioritize patchingCVE-2026-123007CVE-2026-123633CVE-2026-121810+4 moreChecked Nov 18

What it means for your AI agents

Agent backends, MCP servers and webhook receivers are usually built on these runtimes and libraries. A flaw that runs code before authentication hands an attacker the agent's keys and tool access, not just the server.

  • Check the services behind your agents and MCP tools, including ones that only call outward.
  • Look for the affected component in dependencies too, not just the runtime you installed directly.
  • Until patched, treat output from an affected service as untrusted input to your agents.

Updates

  • Nov 18, 2026 : F5 and nginx.org disclosed ten nginx vulnerabilities fixed in nginx 1.30.7 and 1.31.9 , including the five Xint calls nginx2root, and Xint published technical details .

Details

On November 18, 2026, F5 and nginx.org disclosed ten nginx vulnerabilities. Xint, which reported five of them, published technical details on those five under the name nginx2root. Each is an independent path to memory corruption in a worker process, in a different configuration. F5 rates each one High (CVSS 3.1 score 8.1) and Critical (CVSS 4.0 score 9.2), and says code execution requires ASLR to be disabled or bypassed. Xint describes leaking memory addresses to bypass it.

CVE-2026-123007 is an overflow in HTTP/2 flow control that leads to out-of-bounds heap reads and writes. It affects any server with HTTP/2 enabled, from nginx 1.9.5. Xint demonstrated command execution against a public HTTPS endpoint on AWS, GCP and Azure, and notes that the HTTP/2 control frames the attack uses are not logged as requests.

CVE-2026-123633 and CVE-2026-121810 are use-after-free flaws in HTTP/2 (from 1.19.7) and HTTP/3 (from 1.25.0) when nginx proxies requests to a backend and a worker runs low on worker_connections. F5 says exploiting the HTTP/3 flaw also depends on conditions beyond the attacker's control. CVE-2026-123634 is a use-after-free in the resolver when proxy_pass uses variables and an SSI or mirror subrequest resolves the same name. CVE-2026-113425 is a use-after-free in the mail proxy; Xint reports it with auth_http, triggered during login without valid credentials.

Code execution is Xint's claim: it reports remote code execution with worker privileges for each flaw in its tests. Some of its exploits needed more than reaching nginx. For CVE-2026-123633 and CVE-2026-121810, nginx had to be able to open a TCP connection to the attacker, and for CVE-2026-123634, nginx's resolver had to reach the attacker's DNS server.

The sources differ on scope and severity. F5 and nginx.org list the resolver flaw from 0.6.18 and the mail proxy flaw from 0.5.33, while Xint gives 1.13.4 and 0.5.15 as the first releases for the paths it tested. nginx.org rates CVE-2026-123007 and CVE-2026-113425 major and the other three medium. F5 also credits other researchers for CVE-2026-123007 and CVE-2026-123633.

The same releases fix five further nginx vulnerabilities. The most serious, CVE-2026-118899, is an infinite loop while reading a request body over TLS 1.3 that lets an unauthenticated attacker stall a worker process, in nginx 1.13.0 and later. nginx.org rates it major and F5 rates it High (CVSS 3.1 score 7.5). nginx.org rates the other four medium or minor; most affect the stream modules, and one, CVE-2026-120993, is fixed only in 1.31.9.

Xint also reports a use-after-free in the Linux kernel's SCTP send path, present since v4.15, which it chained with CVE-2026-123007 to gain root on the host, including from nginx in a Docker container. It is withholding the details until a kernel fix is available.

As of November 18, none of the five is in CISA's Known Exploited Vulnerabilities catalog. Xint omitted the final exploitation steps, and the sources reviewed do not report exploitation in the wild.

Risk

Any nginx reachable directly from the Internet is at risk. Xint reports running commands as the nginx worker on HTTP/2 servers; F5 says that requires bypassing ASLR.

Affected deployments

NGINX Open Source before 1.30.7 or 1.31.9, and NGINX Plus 37.x before 37.0.8.1 or 37.3.0.1 (37.0.7.1 or 37.2.0.1 for CVE-2026-123633), that use HTTP/2, HTTP/3, proxy_pass with variables together with SSI or mirror, or the mail proxy. F5 lists NGINX Plus R33 and later R releases as affected, with no fixed build.

Distribution packages and container images can carry backported fixes without changing the version nginx -v reports, so check the full package version against your distribution's advisory.

Evidence and sources17 links

Check your exposure with an agent

Paste this into an AI agent that can see your infrastructure. It asks for a read-only check of what's affected, including the systems behind your agents, and a table you can act on. Copying it changes nothing.

Prompt
Assess whether the infrastructure you can access is affected by nginx2root: remote code execution in nginx HTTP/2, HTTP/3, resolver and mail proxy (CVE-2026-123007, CVE-2026-123633, CVE-2026-121810, CVE-2026-123634, CVE-2026-113425, CVE-2026-118899 and CVE-2026-120993). Work read-only within your authorized access: do not run exploits, change configuration, restart services or install anything.

1. Confirm the affected and fixed versions from the vendor advisories and CVE records:
   - https://nginx.org/en/security_advisories.html
   - https://my.f5.com/manage/s/article/K000163925
   - https://my.f5.com/manage/s/article/K000163931
   - https://my.f5.com/manage/s/article/K000164058
   - https://my.f5.com/manage/s/article/K000163926
   - https://my.f5.com/manage/s/article/K000163928
   Cite each source and the date you read it. Say plainly what you could not verify.

2. Inventory the systems that could run the affected software, including the hosts, containers, MCP servers, tool runners and sandboxes behind any AI agents. For each, record the asset, environment, owner if known and the version actually running, from read-only evidence (package metadata, version commands, platform inventory).

3. Note which affected systems are reachable from the internet or from agents and tools, and which credentials (model-provider keys, tool tokens, service accounts) are stored on them.

4. Report one row per asset: asset, running version, affected (yes / no / unknown), reachable from, credentials at risk, evidence, and the next step. List anything you could not access as a gap rather than assuming it is safe.

More from the Threat Wire

Briefing source: Cloudflare Radar Zero Wire (Cloudforce One), last checked November 18, 2026. Refreshed here every 30 minutes.