Critical Red Hat Quay Flaw Lets Attackers Read and Overwrite All Private Container Images in a Registry
Reading a private container image means obtaining its contents even though the repository is not public. Images can contain application files and, as Red Hat warns, secrets or credentials. Overwriting means pushing a changed image in place of the one teams expect. In Quay, an attacker exploiting the flaw can use accepted tokens to pull private images or push replacements. The change can spread beyond the registry: every system that later pulls the replaced image receives the changed version. That makes a registry compromise a supply-chain risk. An attacker might expose embedded credentials, alter software delivered to systems, or disrupt normal image use. The article says an attacker-controlled key can also delete keys used by running Quay instances, causing pushes and pulls to fail with authentication errors. Fixed builds are available, but removing a planted key alone is temporary until the system is updated.
What does it mean for an attacker to read or overwrite a private container image in a Quay registry?
Reading a private container image means obtaining its contents even though the repository is not public. Images can contain application files and, as Red Hat warns, secrets or credentials. Overwriting means pushing a changed image in place of the one teams expect.
In Quay, an attacker exploiting the flaw can use accepted tokens to pull private images or push replacements. The change can spread beyond the registry: every system that later pulls the replaced image receives the changed version. That makes a registry compromise a supply-chain risk.
An attacker might expose embedded credentials, alter software delivered to systems, or disrupt normal image use. The article says an attacker-controlled key can also delete keys used by running Quay instances, causing pushes and pulls to fail with authentication errors. Fixed builds are available, but removing a planted key alone is temporary until the system is updated.
How widespread is the exposure: which Quay versions and deployment types are affected, and does an attacker need an account?
Red Hat lists every version of Quay 3 and Quay 5 as affected by both flaws. For the key flaw, the guidance names standalone installations, Quay deployed by the Quay Operator on OpenShift, and the mirror registry for Red Hat OpenShift. Any instance reachable by untrusted clients is exposed, including internal registries.
For CVE-2026-123710, an attacker needs no account if Quay's default anonymous access is enabled. If anonymous access is off, a valid account still permits exploitation. The second flaw has a different access condition: an attacker must be a Quay user who can create a namespace, but does not need elevated privileges.
Red Hat released fixed builds for eight Quay 3 branches. The article says the CVE page for the first flaw still listed every Quay package as affected, despite the advisories. For the second flaw, the CVE page names eight fixed releases, but none of the eight advisories list that CVE.
What could happen after an attacker steals robot account tokens or replaces an image that other systems later pull?
Robot accounts provide credentials that can be used to access a tenant's images. The second Quay flaw can expose another tenant's robot account tokens, letting an attacker use them to read, push, or overwrite that tenant's images. Losing control of those credentials can therefore extend an attack beyond the registry account that was initially targeted.
The flaw is in the code that lists robot accounts. An unescaped underscore in a namespace name becomes a SQL LIKE wildcard, so a namespace can match another tenant's namespace and return its robot accounts with decrypted tokens. If the stolen credentials are used to replace an image, the changed image can spread to every system that later pulls it.
Red Hat says that, when Quay is OpenShift's integrated registry, stolen robot credentials can inject malicious images into CI/CD pipelines. Deleting the attacker's namespace also deletes its corresponding robot accounts, potentially cutting off systems that rely on them. Updating is the only complete fix for this flaw.
How does adding an attacker-controlled signing key let Quay accept tokens for any repository?
Quay uses trusted keys to verify access tokens for pushing and pulling images. If Quay accepts a token signed by a key, it treats that token as valid for the access the token grants. The key flaw matters because an attacker can add a key they control to the set Quay trusts.
Quay's key server checks that a key-replacement request has a token signed by a trusted key, with the expected audience and issuer. But it does not verify that the token was issued for key rotation. An anonymous token from Quay's /v2/auth endpoint can pass: Quay signs it with a trusted key, and the caller chooses its audience. The planted key has no expiry date.
Quay then accepts tokens signed by that key for any repository, allowing access to private images and image pushes or overwrites. The planted key survives restarts until removed, and deleting it is not enough while the flaw remains. The upstream fix adds a rotation check, requires expiry dates, and prevents callers choosing the audience.
Why should a service check what a token was issued for—not just whether it was signed by a trusted key?
A signature helps a service verify that a token was signed by a key it trusts. But that alone does not establish what the token is meant to authorize. A service also needs to check that the token's purpose and intended recipient match the requested operation. Otherwise, a valid token issued for one context may be misused in another.
Quay's key server checked for a trusted signature and the expected audience and issuer when a caller requested a key change. It did not check whether the token had been issued for key rotation. An anonymous token from Quay's /v2/auth endpoint could pass because Quay signed it with a trusted key and the caller chose its audience.
That gap let an attacker add a key they controlled, then use its tokens for any repository. The upstream fix adds the missing rotation check and stops callers from choosing a token's audience. These checks help keep token validity tied to the operation for which a token was issued.
How can an unescaped underscore in a SQL search pattern cause one tenant’s robot accounts to match another tenant’s namespace?
SQL LIKE uses special characters to match patterns rather than only literal text. In particular, an underscore matches any single character. A search that inserts a namespace name into a LIKE pattern without escaping that underscore can therefore match names that differ at that position.
Quay's robot-account listing code builds a SQL LIKE pattern from a namespace name without escaping wildcard characters. A user who can create a namespace with an underscore in the right place can make the search match another tenant's namespace. The results can include that tenant's robot accounts and their decrypted tokens. The attacker needs no elevated privileges, and the victim does not have to act.
Those stolen tokens can be used to read, push, or overwrite the victim's images. Red Hat's upstream fix also addresses the robot-account search. Until an update is installed, the company suggests limiting creation of organizations or namespaces containing underscores, rotating tokens for matching namespaces, and watching for unusual pulls or pushes. These steps reduce exposure but do not fix the flaw.
What are container registries and images, and why do software teams rely on them to build and run applications?
In general, a container registry is a service for storing and distributing container images. An image is a packaged collection of files and settings used to run software in a container. Teams use registries as a place to keep images and make them available to systems that need to pull them. These are general definitions; the article focuses on Quay and its access flaws.
The article describes teams pushing images to Quay and later pulling them from repositories. Some images are private, and images can contain secrets or credentials. Robot accounts also provide tokens used to read, push, or overwrite tenant images. Those details show why teams need a reliable way to control who can access and change registry contents.
When an image is overwritten, the change can reach every system that later pulls it. Red Hat also warns that stolen robot credentials can inject malicious images into CI/CD pipelines when Quay is OpenShift's integrated registry. The article recommends fixed releases and temporary access controls while updates are pending.
This brief was written by AI from the original reporting and checked by other models. Names, figures and quotes come from the source; read it for full context.
Read more in the JupiteX app
Pulse is free. New stories every 4 hours, each one broken into the questions that explain it.
Or read more news on the web