Red Hat has disclosed two critical vulnerabilities in the Red Hat Quay container registry, each of which allows an attacker to gain full access to other users’ private container images. CVE-2026-123710 (CVSS 9.8) makes it possible to inject an attacker’s own signing key and read or overwrite images in all repositories — when anonymous access is enabled, this does not even require an account. CVE-2026-110697 (CVSS 9.9) allows an attacker, through SQL injection in the service account search mechanism, to steal tokens belonging to other tenants. Both vulnerabilities were discovered by an internal Red Hat researcher, and patches have already been released for eight Quay 3 branches. Organizations using Quay in any configuration — standalone, via the operator on OpenShift, or as a mirror registry — need to update immediately.
CVE-2026-123710: registry signing key takeover
According to Red Hat’s advisory, the vulnerability is related to insufficient input validation (CWE-1287) in the Quay key server. Quay uses trusted keys to sign access tokens that authorize push and pull operations on images. The vulnerability enables an attacker to add their own key to the trusted list — without an expiration date. After that, tokens signed with this key are accepted for any repository.
The root cause is that the key server checks that a key rotation request contains a token signed by a trusted key, with the expected audience and issuer. However, it does not check whether the token was actually issued for the key rotation operation. An anonymous token obtained through the /v2/auth endpoint passes this check, because Quay signs it with a trusted key, and the caller is free to choose the audience.
Consequences of exploitation:
- Reading private images and any secrets or credentials contained within them
- Overwriting images, with the changes propagating to all systems that subsequently pull those images
- Deleting keys used by running Quay instances, leading to denial of service — normal push and pull operations start failing with authentication errors
The injected key persists across restarts until it is removed manually. According to the bug report, the vulnerable logic has been present in Quay since the first public commit.
A crucial nuance: according to the Red Hat mitigation guide, disabling anonymous access does not protect against exploitation. Anonymous access only enables the attack without an account; when it is disabled, any user with a valid account can still exploit the vulnerability. The injected key also continues to work after password resets and rotation of service account tokens.
CVE-2026-110697: token leakage via SQL injection
The second vulnerability, described on the Red Hat CVE page, is classified as CWE-89 (SQL injection) with a CVSS score of 9.9. It is located in the code responsible for listing service accounts (robot accounts). When building the SQL LIKE pattern from the namespace name, wildcard characters are not escaped. In SQL LIKE, the underscore character matches any single character.
This means that an attacker who creates a namespace with an underscore in the right position can cause it to match another tenant’s namespace and obtain that tenant’s service accounts along with their decrypted tokens. Exploitation does not require elevated privileges — it is enough to be able to create a namespace. The victim does not need to perform any actions.
The stolen tokens allow the attacker to read, upload, and overwrite the target tenant’s images. Red Hat notes that in environments where Quay serves as the integrated registry for OpenShift, stolen service account credentials can be used to inject malicious images into CI/CD pipelines.
A public repository has been found for this vulnerability, containing an alleged proof of concept (PoC). However, no confirmed cases of exploitation in real-world environments have been identified so far. Neither vulnerability is currently listed in the CISA KEV catalog.
Differences in attack prerequisites
The two vulnerabilities have different requirements for the attacker, which is important for prioritizing defensive measures. CVE-2026-123710 can be exploited without authentication if the /keys endpoint is accessible to untrusted clients. CVE-2026-110697 requires an authenticated user with permissions to create a namespace and the presence of a “collision” via the underscore character. Accordingly, defenders need to assess both the exposure of key endpoints and the control over creating namespaces and organizations.
Affected versions and patches
Red Hat lists the following versions as containing fixes for both vulnerabilities:
- Quay 3.9.28
- Quay 3.10.28
- Quay 3.12.24
- Quay 3.14.10
- Quay 3.15.9
- Quay 3.16.7
- Quay 3.17.7
- Quay 3.18.3
At the same time, Quay 5 and Mirror Registry for Red Hat OpenShift 2 are still listed as affected on the CVE-2026-123710 page. According to Red Hat, the Quay.io cloud service has already been patched for CVE-2026-110697.
The fix in the upstream repository adds verification of key rotation, requires keys added via the key server to have an expiration date, forbids the caller from choosing the token audience, and escapes namespace prefixes when searching for service accounts.
Mitigation recommendations
For CVE-2026-123710, Red Hat recommends the following sequence of actions:
- At the proxy or load balancer level, block the /keys endpoint for everything except trusted internal components
- Check audit logs for service_key_rotate and service_key_delete events where expiration_date is null
- Delete all unrecognized approved service keys
- Install a patched version — until then, an attacker can re-add a key using the same method, so removing the key without updating is not a durable solution
For CVE-2026-110697, updating is the only complete solution. Interim measures only reduce risk:
- Restrict the ability to create organizations and namespaces containing the underscore character
- Rotate service account tokens for namespaces that may collide with namespaces containing an underscore
- Monitor pull and push operations that do not match the usual behavior of service accounts
We have previously covered critical vulnerabilities in products affecting the Red Hat ecosystem — in particular, RCE in the Linux kernel IPv4 TCP.
The priority for security teams is clear: deploying the patch must come before all other measures. Removing compromised keys without updating Quay does not provide protection, because an attacker can re-inject a key through the same mechanism. Organizations using Quay in any deployment configuration should update to one of the listed patched versions and then audit keys and service account tokens for signs of compromise.












