Introduction
An unauthenticated client that can reach a Red Hat Quay instance with anonymous access enabled may be able to install its own permanently trusted signing key, then mint registry tokens that Quay honors for every repository. For organizations that use Quay to hold private container images, that translates into potential disclosure of private images and unauthorized replacement of the tags that build and deployment systems pull from.
Red Hat Quay is a private container image registry that Red Hat describes as distributed and highly available, offered both as a self-managed product and as the hosted Quay.io service. Red Hat positions it as one of the early enterprise hosted and on-premises registries and markets it for use with Red Hat OpenShift and Kubernetes. The materials we reviewed contain no independent market share or adoption figures. Because a registry sits upstream of every cluster that pulls from it, a failure in its token-signing trust chain affects software supply chain integrity as well as the registry itself.
Red Hat rates CVE 2026 123710 at CVSS 3.1 9.8 (Critical): network attack vector, low attack complexity, no privileges required, no user interaction, and high confidentiality, integrity, and availability impact. NVD has not published its own assessment and marks the record as awaiting enrichment. At the time of writing, the reviewed sources do not document exploitation in the wild, a public exploit, or any named threat actor.
Technical information
Root cause
The weakness is classified as CWE 1287, Improper Validation of Specified Type of Input. The underlying problem is token purpose confusion in Quay's service key rotation endpoint, compounded by two permissive behaviors in how rotated keys are stored:
- The key rotation service checks that the submitted token has an accepted signature, audience, and issuer, but not that the token was issued for key rotation. An anonymous registry token, which Quay will hand to anyone when anonymous access is on, passes those checks. Its audience comes from the client-supplied
serviceparameter, so the attacker controls that value. replace_service_keycopies the signer's approval onto the newly submitted key, so the attacker's key lands in the trusted set already approved.- If the expiration parameter is omitted, the new key is stored with no expiration at all.
In effect, Quay accepts a credential meant for anonymous registry access as authorization to modify its own signing key set. Once an attacker key is approved, Quay trusts registry tokens signed with it for any repository, which crosses the boundary between registry authentication and per-repository access control.
Attack flow
Based on Red Hat Bugzilla 2568072, the attack proceeds as follows:
- The attacker sends an anonymous request to
/v2/auth, which only works whenFEATURE_ANONYMOUS_ACCESSis enabled. The request carries aserviceparameter of the attacker's choosing, and that value becomes the token's audience. - Using that token as a bearer JWT, the attacker sends a PUT containing their own JWK to
/keys/services/quay/keys/<kid>. This requires/keysto be reachable by the attacker. replace_service_keycopies the signer's approval onto the new key. With the expiration parameter left out, the key is stored with no expiration.- The rotation deletes the legitimate instance key as a side effect. From then on, the attacker's key can delete any replacement instance keys.
- The attacker signs registry JWTs with their key and uses them against
/v2to pull private images, push images, and overwrite tags in any repository.
Why cleanup alone is not enough
The Bugzilla report notes that deleting the rogue key is not durable before a fixed build is deployed. The preconditions that allowed the first rotation are still present, so another anonymous request can install a new key. Responders should expect that a key removed in the morning could be replaced by the afternoon unless /keys exposure is closed.
The issue was confirmed against the Quay master revision and a 3.19 unstable image. Those were test targets and do not define the full affected version range.
Exposure implications
A single trusted signing key authorizes operations across repositories, so affected organizations should treat this as potential private image disclosure plus unauthorized image or tag modification across the whole registry, rather than a permission error scoped to one repository. Tag replacement in particular means downstream consumers could pull an image digest the repository owner never pushed.
Detection methods
No public YARA, Sigma, Snort or Suricata signatures, and no atomic IoCs (IPs, hashes, domains), exist for CVE 2026 123710 yet. Detection relies on vendor guidance from Red Hat and on the exploit mechanics in the Bugzilla report. That works reasonably well here, because a successful attack has to change Quay's trusted service key set and so leaves traces in both the audit trail and the key inventory.
Primary method: hunt the Quay audit log
Red Hat's advisory says to review audit logs for service_key_rotate and service_key_delete events with a null expiration_date, and to remove any approved service keys you do not recognize. This catches the key installation itself (step 3), not the earlier token request. Legitimate rotations by Quay's own components normally carry an expiry, so a rotation without one stands out. Look for:
service_key_rotateevents with a nullexpiration_date. This is the clearest sign that a key with no expiry was installed through this path.service_key_deleteevents that do not match planned maintenance, especially shortly after a rotation. Rotation deletes the real instance key, and the attacker's key can delete its replacements.- Repeated rotate and delete cycles. Because the attacker can repeat the anonymous request, a rotate event shortly after you removed a rogue key may indicate renewed targeting rather than a false positive on the first finding. Keep monitoring after cleanup instead of treating one sweep as the end of the investigation.
Secondary method: check the service key inventory
Compare the approved service keys for the quay service against what you expect. Your own Quay instances and trusted internal components should be the only owners. Red flags:
- an approved key with no expiration
- a
kidthat cannot be linked to a known Quay instance - a running Quay instance whose own key has disappeared
The advisory directs administrators to remove unrecognized approved keys, so this inventory check is part of the vendor's recommended response.
Supporting telemetry: proxy and load balancer logs
The Bugzilla report names the exact endpoints, so HTTP logs in front of Quay are useful. These are hunting ideas derived from that report, not published signatures:
- PUT requests to
/keys/services/quay/keys/<kid>from outside your trusted internal components. Red Hat notes that exposure depends on whether/keysis reachable by untrusted clients, so any external request here deserves investigation. - Anonymous
/v2/authrequests with an unusualserviceparameter, especially when followed soon after by a/keysrequest from the same client. - Pushes or tag changes on
/v2that cannot be attributed to a known user or robot account, particularly after a suspicious key event. Forged tokens grant push and overwrite rights on any repository, so review recent tag changes on sensitive repositories for unexpected image digests.
Exposure check
Two conditions from the advisory and the Bugzilla report make an instance exploitable:
- Anonymous access is enabled. Check whether your configuration sets
FEATURE_ANONYMOUS_ACCESS. /keysis reachable from untrusted networks. Probe it from an external vantage point.
Instances meeting both conditions should get the audit log and key inventory hunts above first.
Affected systems and versions
- Products listed by NVD: Red Hat Quay 3 and Red Hat Quay 5. NVD supplies no version ranges or version boundaries.
- Confirmed test targets: the Quay master revision and a
3.19 unstableimage. These confirm the flaw exists but do not establish a complete affected range. - Vulnerable configuration:
FEATURE_ANONYMOUS_ACCESSenabled (required for anonymous token issuance) and the/keysendpoint reachable by untrusted clients. - Fix status: the Bugzilla record was marked NEW when last modified on November 11, 2026, with a blank fixed version field. Neither the Red Hat advisory page nor the Bugzilla report reviewed here identifies a fixed release, so we cannot name an upgrade target.
Red Hat's documented interim mitigation is to restrict /keys at the proxy or load balancer so only trusted internal components can reach it, then review the audit log and remove unrecognized approved keys as described above. Treat key removal as containment, not resolution, and check the official Red Hat advisory for current release guidance before making deployment changes.
Vendor security history
The evidence supports only a limited observation. NVD records a separate 2026 Quay flaw, CVE 2026 32589, that could let an authenticated user with push permission interfere with image uploads in other repositories. That issue and this one are different failure modes, though both involve actions escaping their intended repository scope. The available sources do not establish how often such issues occur relative to other registry products, and they are not sufficient to rate Red Hat's overall security performance.
References
- NVD: CVE 2026 123710
- CVE Record: CVE 2026 123710
- Red Hat Customer Portal: CVE 2026 123710 advisory
- Red Hat Bugzilla 2568072: unauthenticated registry signing key takeover via service key rotation
- StackFlag: Red Hat Quay lets unauthorized users add permanent signing key
- Red Hat Quay product page
- Red Hat Quay datasheet
- NVD: CVE 2026 32589 (separate Quay flaw)



