Two flaws disclosed on November 11, 2026, scored 9.8 and 9.9, can turn a reachable Quay registry into a path to private images in every repository. One bypasses the registry’s trust boundary by installing an attacker-controlled signing key; the other crosses tenant boundaries by returning another organization’s robot accounts and decrypted tokens.
In this article
Why this matters
A container registry is not merely a storage layer. It is a trust anchor: images pulled from it are treated as authoritative by deployment pipelines, clusters and operators. The two Quay flaws attack that trust in different ways. CVE-2026-123710 lets an attacker mint registry tokens that Quay accepts for any repository, including private ones. CVE-2026-110697 lets a low-privilege user retrieve robot-account tokens belonging to another tenant and reuse those credentials to read, push or overwrite that tenant’s images.
The practical consequence is supply-chain exposure. An attacker who can overwrite an image can place malicious content where downstream systems expect a trusted artifact, and the change propagates to every system that subsequently pulls that tag. An attacker who can pull private images may also obtain credentials or secrets embedded in image layers. Red Hat additionally notes that, where Quay serves as the integrated registry for OpenShift, stolen robot credentials can be used to inject malicious images into CI/CD pipelines (Red Hat CVE-2026-110697).
The failed assumption is not simply that anonymous access was enabled or that a namespace name was validated. In the first flaw, Quay assumed that a token signed by a trusted key and carrying the expected audience and issuer necessarily authorized key rotation. In the second, it assumed that a namespace string could be embedded in a SQL LIKE pattern without escaping wildcard characters. Both assumptions allowed ordinary registry-facing inputs to cross a boundary that operators reasonably expect to hold.
| Claim | Status | Basis |
|---|---|---|
| An unauthenticated caller can install an approved signing key with no expiry | Confirmed | Red Hat’s CVE statement and Bugzilla 2568072 |
| Tokens signed with the installed key are accepted for any repository | Confirmed | Red Hat’s CVE statement and remediation guidance |
| A namespace containing an underscore can match another tenant’s namespace | Confirmed | Red Hat’s CVE statement and Bugzilla 2552237 |
| Robot-account listings expose decrypted authentication tokens | Confirmed | Red Hat’s CVE statement and remediation guidance |
| The flaws have been exploited in the wild | Not established | No public exploitation evidence identified in the record |
| Quay.io was already fixed for CVE-2026-110697 at disclosure | Company statement | Red Hat remediation solution 7149592 |
What happened
Red Hat disclosed CVE-2026-123710 and CVE-2026-110697 on November 11, 2026. The first is an unauthenticated service signing-key takeover in Quay’s key server; the second is a SQL LIKE wildcard injection in robot-account listing. Red Hat assigned CVSS v3.1 base scores of 9.8 and 9.9 respectively, with CWE-1287 and CWE-89 as the assigned weakness types. Red Hat is the CVE Naming Authority for these records, and NVD had not yet published its own base scores at the time of the record (Red Hat CVE-2026-123710).
For CVE-2026-123710, the attack begins at /v2/auth. Quay can mint an anonymous registry token for a caller, and the caller controls the token’s audience through the service parameter. When that token reaches the service key rotation endpoint, the key server verifies the signature, audience and issuer but does not check that the token was issued for key rotation. The rotation operation then copies approval onto the attacker’s new key and, when the expiration parameter is omitted, stores it without an expiry date. Registry tokens signed with that key are accepted for any repository (Red Hat Bugzilla 2568072).
The consequences extend beyond image theft. An attacker can pull private images, push or overwrite them, and replace repository tags. The planted key can survive restarts and is not revoked by password resets or robot-token regeneration. Rotation also deletes the legitimate instance key as a side effect, and the attacker’s key can delete replacement keys, causing legitimate clients to receive authentication or authorization errors (Red Hat solution 7149923).
Anonymous access is enabled by default in Quay, so no account is required in the default configuration. Disabling it does not close the flaw: any valid Quay account can exploit it. Red Hat states that deployments where /keys is reachable by untrusted clients are affected, including internal registries, standalone installations, Quay Operator deployments on OpenShift and the Red Hat OpenShift mirror registry (The Hacker News).
CVE-2026-110697 follows a different path. The list_entity_robots function builds a SQL LIKE pattern from a user-controlled namespace without escaping wildcard metacharacters. Because an underscore matches any single character in SQL LIKE, an attacker who can create a namespace can choose a name that matches a victim tenant’s namespace. The resulting listing returns the victim’s robot accounts together with their decrypted authentication tokens (Red Hat Bugzilla 2552237).
Those tokens permit reading, pushing or overwriting the victim’s images. The attacker can also rename the namespace to retarget robot prefixes, or delete the namespace, which deletes the corresponding robot accounts and can disrupt systems that depend on them. No elevated privileges or victim interaction are required; the stated prerequisite is an authenticated user with namespace-creation capability (Red Hat solution 7149592).
The fixes and their limits
Red Hat released fixed builds for eight Quay 3 branches: 3.9.28, 3.10.28, 3.12.24, 3.14.10, 3.15.9, 3.16.7, 3.17.7 and 3.18.3. Quay 5 is listed as affected. The upstream remediation was merged as commit 094cfb024f01 and cherry-picked through pull request #8057 to the redhat-5.0 branch, both on November 11, 2026. RHSA-2026:99019 covers Quay 3.18.3 (GitHub commit 094cfb024f01; RHSA-2026:99019).
For the signing-key flaw, the upstream change scopes service-key lifecycle operations to a specific service. The replacement and deletion functions now require a service parameter, and key selection and deletion add a service condition. The Hacker News also reports that the change requires keys added through the key server to carry an expiry date and stops callers from choosing a token’s audience. The robot-account path is addressed by escaping namespace prefixes before they are used in the search pattern (The Hacker News).
Configuration mitigations can reduce exposure but are not durable substitutes for patching. Red Hat recommends restricting /keys at the proxy or load balancer to trusted internal components, reviewing audit logs for service_key_rotate and service_key_delete events with a null expiration_date, and removing unrecognised approved service keys. Deleting a rogue key does not last while the vulnerable build remains in service, because the same anonymous request can install another one (Red Hat CVE-2026-123710).
For the robot-token flaw, Red Hat advises limiting who can create organizations or namespaces containing underscores, rotating robot tokens for namespaces that match an underscore namespace, and watching for pulls or pushes inconsistent with normal robot use. Red Hat nevertheless states that updating is the only complete fix. Quay.io was already fixed for CVE-2026-110697 at disclosure, although The Hacker News reported that none of the eight Quay 3 advisories listed that CVE (The Hacker News).
What is known, and what is not
The record establishes the technical mechanics, affected product lines, fixed releases and vendor-recommended mitigations. It does not establish that either flaw has been exploited in the wild. The research report states that CISA’s Known Exploited Vulnerabilities catalog, version 2026.11.10, listed neither CVE, and that EPSS percentiles were low: 37th for CVE-2026-123710 and 43rd for CVE-2026-110697. Those indicators describe public threat activity and prioritization, not the absence of compromise (Brandefense).
Red Hat states that the flawed key-trust logic behind CVE-2026-123710 has existed since Quay’s first public commit, making it a multi-year latent defect rather than a recent regression. That history is a vendor statement, not an independently established timeline. Both flaws were discovered by Red Hat’s Dave O'Connor (Red Hat Bugzilla 2568072).
The operational priority is therefore clear even without evidence of active exploitation: patch affected registries, then determine whether unauthorized service keys or cross-tenant robot activity already exist. Because a planted key can survive restarts and credential resets, and because deleting it does not prevent reinstallation on a vulnerable build, remediation must combine investigation with deployment of a fixed release.
Method and sources
This analysis relies on Red Hat’s CVE records and remediation solutions, Red Hat Bugzilla reports, the upstream Quay commit and pull request, the RHSA erratum, GitHub advisories, NVD and cve.org records, and reporting from The Hacker News, SecurityWeek, Aviatrix and Brandefense. Red Hat is treated as the authoritative CNA for its products; NVD’s scores were still listed as N/A in the record.
Claims are separated by evidentiary strength: confirmed for vendor-documented mechanics and released fixes, company statement for assertions attributed to Red Hat but not independently verified, and not established where the record contains no evidence of exploitation. No independent compromise assessment, packet capture or registry audit was available, so the absence of reported exploitation should not be read as proof that no deployment has been affected.