On this page (7 sections)

Miren Runtime, the runtime component of the Miren application platform, has a critical authentication bypass in its RPC server. According to the GitHub advisory GHSA-8fh7-7q4q-cq52, versions before 0.11.1 request a TLS client certificate but never check that it was issued by the cluster certificate authority, so a forged certificate is accepted and treated as a cluster superuser.

If you run Miren Runtime anywhere, especially with the RPC port reachable from outside a tightly controlled network, you should upgrade to 0.11.1, restart the server, and follow the vendor's advice to redeploy applications and rotate secrets. So far the only source we found is the maintainers' own advisory, so details below come from that single primary source.

What happened

The Miren Runtime RPC server listens on port 8443 using QUIC/HTTP3. It is the control plane interface for the cluster: through it a caller can manage apps, sandboxes, disks and other resources. The advisory says the server asked clients for a certificate but did not verify it against the cluster CA, and that any identity arriving via a client certificate was then allowed past all role-based access checks.

The result is a complete, unauthenticated bypass. The advisory states that no valid CA signature, credential or prior authorization is needed, and that an attacker with network access to the RPC port can perform any RPC operation with full privileges. The maintainers describe this as full compromise of confidentiality, integrity and availability of the cluster and everything it orchestrates.

Why it worked: three flaws combined

This is a good example of a common weakness class: a system that checks whether a certificate is present, but not whether it is genuine. The advisory lists three independent factors that combined:

  • The listener requested a client certificate but did not verify the chain against the configured CA, so the verified-chains list was always empty and no later hook re-checked it.
  • The authenticators took the caller's identity from the certificate's common name without confirming that the chain had been verified.
  • The authorizer allowed every identity that came from a certificate and skipped RBAC entirely.

Any one of these controls, working correctly, would have stopped the attack. Together they meant that the contents of an unverified certificate were trusted as proof of who the caller was. The advisory also notes that the client-side check only compares the public issuer name, not a signature, which is why knowing the cluster CA's public subject was enough to get a client to present a certificate. We are deliberately not describing how to build one.

Who is exposed

Exposure depends on network reachability alone, per the advisory. There are two realistic situations:

  • Port 8443 open to the internet (for example ingress from 0.0.0.0/0): anyone can reach the listener, so the flaw is remotely exploitable by anyone.
  • Port 8443 limited to a trusted network: any party who can get onto that network, such as a compromised host, a malicious insider or a neighboring workload, could exploit it.

Because the endpoint is the cluster control plane, the blast radius includes the applications Miren manages and any credentials those applications hold. If you host customer websites or APIs on a Miren cluster, treat them as in scope.

How to check your exposure

First confirm which version you run. Check your deployment tooling or release notes for the installed Miren Runtime version; anything below 0.11.1 is affected. Then see who can reach the RPC port.

This read-only command lists UDP listeners (QUIC uses UDP) on a Linux host and filters for port 8443. Run it on the Miren host; a match means the RPC listener is bound there, and the address column shows whether it is bound to all interfaces or only a private one.

List listeners on port 8443 (Linux, run on the Miren host)Shell
ss -ulnp | grep 8443

Then review your cloud security groups, firewall rules and load balancer settings for any rule that allows inbound traffic to 8443 from broad ranges. The listener binding alone does not tell you who can reach it from outside.

What to do now

  1. Upgrade Miren Runtime to 0.11.1 or later, and restart the server so the new listener configuration takes effect.
  2. Until then, remove any 0.0.0.0/0 ingress to port 8443 and limit it to trusted management networks. The advisory calls this containment only; it does not remove the vulnerability.
  3. Redeploy all applications, as Miren recommends, because unauthorized RPC access cannot be detected after the fact.
  4. Rotate every credential referenced in environment variables of your applications.
  5. Run the maintainers' read-only audit script to review standing access, recent changes and the exact secrets to rotate (linked from the advisory).
  6. Review who and what can reach the cluster network, and tighten it so the control plane is never on a shared or public segment.
Caution. Restarting the server and redeploying applications can interrupt running workloads. Plan a maintenance window, and rotate secrets in an order that avoids breaking dependent services.

The fix verifies the client-certificate chain before trusting it. Per the advisory, the listener now requires a presented certificate to chain to the configured CA, while callers that authenticate with JWT or OIDC through the Authorization header are still admitted. As defense in depth, both authenticators now require a verified chain before deriving an identity from a certificate.

The advisory says Miren cannot detect whether the RPC was accessed without authorization. That is why the vendor recommends redeploying and rotating rather than only patching. If your cluster was reachable from the internet before you upgraded, assume secrets in environment variables may have been exposed.

Detection and monitoring afterwards

Since the product itself cannot show unauthorized RPC use, rely on surrounding evidence. Look at changes to apps, sandboxes and disks that nobody on your team made, unexpected new deployments, and network logs showing connections to port 8443 from addresses outside your management ranges. Review the usage of the credentials you rotated at the services that issued them, such as cloud providers, databases and payment or email APIs, for activity dated before the rotation.

If the cluster serves public websites, also check that the sites themselves have not been altered. Our guide on how to check if a website is hacked and the steps in website hacked: what to do cover that side. Externally visible changes such as new certificates, DNS records or response headers can be watched with Webforta Monitor, which helps you notice that something on the public surface moved, though it does not see inside your cluster.

Timeline and sourcing

Dates as given in the GitHub advisory
DateEvent
Jul 8, 2026Advisory published in the mirendev/runtime repository
Nov 18, 2026Published by the National Vulnerability Database
Nov 19, 2026Published to and reviewed in the GitHub Advisory Database

The fix is in release v0.11.1, with the code change in this commit. The NVD entry for CVE-2026-75521 carries the same CVSS 10 rating. The research shows no independent analysis or reports of exploitation yet, so treat the technical description as the maintainers' account.

Tip. Make a habit of keeping orchestration and management ports off public networks entirely. For general guidance see the small business website security guide.

Sources

This article is based on the public sources below. Details can change as vendors and researchers publish more; check the original advisories before acting on version numbers.

  1. CVE-2026-75521 - GitHub Advisory Databasegithub.com
  • Miren Runtime
  • CVE-2026-75521
  • authentication bypass
  • TLS
  • cluster security
  • Go