The image-customization-controller used in the OpenShift bare metal provisioning stack runs an unauthenticated HTTP image server on port 8084 that lists all served images in a browsable directory. An adjacent attacker on the provisioning network can obtain the image UUIDs, download the images and extract the embedded Ignition configuration. This can expose the cluster pull secret and, if configured, the RAMDISK SSH key.
Red Hat · OpenShift Container Platform
Konflux CI automation scripts store GitHub access tokens in plaintext on the CI runner and do not remove them after execution. A local attacker with code execution or access to the CI environment can read these tokens and gain unauthorized repository access, potentially modifying source code or release artifacts. Red Hat rates the issue as Important (CVSS 3.1: 8.6), while VulDB lists a lower score; exploitation requires prior access to the pipeline runner.
Red Hat · Konflux CI
The containerized OpenShift installer downloads inventory and dynamic scripts from remote URLs and runs them without verifying their integrity. An attacker with a Man-in-the-Middle position on the network, or control of the script server, can substitute a malicious executable. It then runs with the installer's user ID and full cluster-provisioning privileges, which allows arbitrary code execution. The risk only applies where INVENTORY_URL or DYNAMIC_SCRIPT_URL is used.
Red Hat · openshift-ansible
The IIOP listener of Red Hat JBoss Enterprise Application Platform (EAP), based on WildFly, may deserialize input streams before authentication is performed. An attacker who can reach the IIOP port may send crafted serialized objects and achieve code execution if an exposed EJB accepts untyped method parameters and suitable gadget chains exist on the classpath. Red Hat rates the issue as Moderate because standard deployments bind IIOP to localhost by default.
Red Hat · JBoss Enterprise Application Platform
A flaw in the IIOP/CSIv2 security interceptor of WildFly and Red Hat JBoss Enterprise Application Platform lets a remote attacker bypass authentication on IIOP EJB invocations. The interceptor accepts identity assertions (ITTPrincipalName) in the service context without verifying transport-level authentication. An unauthenticated attacker reaching port 3528 could impersonate any principal on IIOP-exposed remote EJBs. Red Hat states that the vulnerability only applies when caller propagation is enabled without mutual TLS or mandatory authentication.
Red Hat · WildFly / JBoss Enterprise Application Platform
A flaw in KServe, shipped with Red Hat OpenShift AI (RHOAI), lets a user with create or update rights on InferenceService resources supply a malicious storageUri. The storage-initializer init container then sends arbitrary outbound HTTP/HTTPS requests from inside the cluster (Server-Side Request Forgery). An attacker can disclose information, including credentials of the node's cloud IAM role, or reach internal cluster services. Exploitation requires a valid, namespace-scoped account.
Red Hat · OpenShift AI (RHOAI)
A flaw in the ai-gateway-payload-processing component of Red Hat OpenShift AI (RHOAI) lets an authenticated user bypass tenant boundaries. By specifying another tenant's model identifier in the request body or header, the attacker causes the gateway to inject that tenant's provider API keys into the request. The attacker can then make unauthorized LLM calls and consume the other tenant's resources and quotas. The flaw affects multi-tenant model serving setups; Red Hat rates it Important, and the CVSS score is 9.1 (Critical).
Red Hat · OpenShift AI
The openjdk-orb component of Red Hat JBoss Enterprise Application Platform (EAP) 7 and 8 accepts attacker-supplied CDR codebase URLs during object unmarshalling on TCP port 3528 (IIOP). This applies only when EAP runs with the Java Security Manager (-secmgr) and the IIOP service is reachable from the attacker's network. An unauthenticated remote attacker can load and instantiate arbitrary classes from a remote URL in the server JVM before the EJB security interceptors run, which Red Hat rates as Important and states can lead to remote code execution. Red Hat describes this issue as an incomplete fix for CVE-2026-15560.
Red Hat · JBoss Enterprise Application Platform
The iscsiuio component of open-iscsi, used for iSCSI with hardware offload NICs on Red Hat Enterprise Linux, miscalculates the length of DHCPv4 option data and reads beyond a heap buffer. An attacker on the same Layer 2 network segment can send a crafted DHCP response to crash the daemon or corrupt its network configuration. According to Red Hat, systems without iSCSI offload hardware are not affected.
Red Hat · open-iscsi (iscsiuio)
When the terminal integration of Goose is enabled, the tool captures shell command history and sends it unfiltered to the configured Large Language Model (LLM) provider during queries. Credentials, API tokens or environment variables typed into the shell can therefore be disclosed to external service providers and to their logs. Exploitation requires that the feature is enabled and that a user takes action, and there is no inbound network service to attack.
Red Hat · Goose
The operator bundle of Red Hat Multicluster Global Hub grants its service account cluster-wide impersonation of arbitrary users, groups and service accounts without restriction. An attacker who has compromised the operator or obtained its service account credentials can impersonate privileged identities, such as system:masters, and gain full administrative control over the cluster. The attacker needs high privileges to reach the operator or its credentials first.
Red Hat · Multicluster Global Hub Operator
Cruise Control, the Kafka cluster management component shipped in Red Hat Streams for Apache Kafka 2 and 3, can be tricked into skipping its authentication checks by adding an extra path segment to a request. The servlet then routes the request to its administrative endpoints, so an unauthenticated caller can run sensitive cluster operations such as rebalancing. The attacker needs network access to Cruise Control port 9090, and Red Hat rates the impact as Important because the Strimzi operator's NetworkPolicy normally restricts that port to operator pods.
Red Hat · Streams for Apache Kafka (Cruise Control)
StreamsHub Console, packaged by Red Hat as console-operator in Streams for Apache Kafka, lets an authenticated Kubernetes tenant with create or update rights on Console custom resources run an arbitrary container image. That image runs under a service account with cluster-wide read and some write permissions on Strimzi resources, so an attacker can read the projected token and access Kafka control-plane resources across tenants. The attack requires a shared cluster where non-administrators can manage Console resources; single-owner clusters have limited exposure.
Red Hat · StreamsHub Console for Apache Kafka
The StreamsHub Console operator (console-operator, Streams for Apache Kafka) can be abused by a Kubernetes tenant who is allowed to create or update Console custom resources. By setting kafkaUser.namespace to another namespace, the attacker makes the operator read KafkaUser secrets there, such as SASL credentials or TLS private keys, and write them in plaintext into a secret in the attacker's own namespace. The attack requires valid low-privilege access to the cluster and no user interaction; the flaw has no same-namespace check. Impact is confidentiality only (CVSS 7.7 from the CNA).
Red Hat · StreamsHub Console
Ghostscript as shipped with Red Hat Enterprise Linux 8 lacks the fix for CVE-2019-14869 after a rebase from version 9.25 to 9.27. A specially crafted PostScript file can bypass the -dSAFER restrictions, allowing access to files outside the permitted paths or command execution with the rights of the Ghostscript process. Exploitation requires that an attacker can get such a file processed, for example by document conversion or print services that handle untrusted input.
Red Hat · Enterprise Linux 8 Ghostscript
Red Hat Advanced Cluster Security for Kubernetes (RHACS) Central does not record SAML assertion or response identifiers that have already been used. An attacker who obtains a valid, unexpired signed SAML response can submit it repeatedly and receive new Central authentication tokens for the mapped identity. Exploitation requires access to such a SAML response and network reachability of the Central SAML endpoint.
Red Hat · Advanced Cluster Security for Kubernetes
A flaw in bootc and ostree lets already-staged OS deployments be finalized without checking their container image against the current signature policy. An attacker with local access who can influence staging or the signature policy can keep unsigned content bootable after the policy is hardened, which can lead to full host compromise on reboot. The attack complexity is high and no privileges or user interaction are required.
Red Hat · bootc and ostree
A flaw in the GLZ dictionary restore path of spice-server truncates a size_t allocation size to int during VM live migration. An authenticated SPICE client can send crafted migration data with a manipulated max_encoders value to trigger a heap buffer overflow. Successful exploitation requires valid SPICE client credentials and an active live migration, and can lead to a crash or code execution in the SPICE server process.
Red Hat · spice-server
A flaw in the WildFly management API allows an authenticated user with the Monitor role to exploit a path traversal in the take-snapshot operation by manipulating the name parameter. The user can write a copy of the server configuration to an arbitrary location and read it back, which exposes sensitive credentials that RBAC would otherwise restrict. The attack requires a valid low-privileged account, and the vendor rates the impact as Low, while NVD/CNA CVSS is 7.1 (High).
Red Hat · WildFly
The Kuadrant operator shipped in Red Hat Connectivity Link and Red Hat OpenShift AI does not correctly escape values from HTTPRoute path, header and query match fields when it generates the CEL predicate. A user who can create or modify an HTTPRoute attached to a shared Gateway can craft a match value that makes the predicate skip Gateway-level authentication and rate limiting. The request is still forwarded to the route's backend. The attack needs authenticated, route-write access to the cluster.
Red Hat · Connectivity Link
A path traversal flaw (CWE-22) in the wildfly-core library, which WildFly and JBoss EAP use to manage deployments, lets an attacker with the Deployer role access files outside the deployment directory. The attacker can delete arbitrary directories or write files with arbitrary content, such as replacing the management credential store, and so gain the management SuperUser role. Exploitation requires a valid account with the Deployer role, but the impact is total.
Red Hat · WildFly Core
Ansible Operator Plugins, used in Red Hat operators built with Ansible, contains a Jinja2 template injection flaw. A low-privileged Kubernetes user who can submit crafted custom resources can make Ansible evaluate embedded Jinja2 expressions inside the operator pod. This can lead to arbitrary code execution and privilege escalation, giving control over the operator's service account, which typically holds cluster-wide permissions. The CVSS vector requires a high-complexity condition, and the scope change means the impact extends beyond the operator component.
Red Hat · Ansible Operator Plugins
A flaw in the VDUSE handler of DPDK lib/vhost fails to invalidate vring pointers when the IOTLB is invalidated, which leads to a use-after-free (CWE-416). A malicious virtual machine with a VDUSE device can trigger it through memory hot-remove, balloon inflation or an IOMMU mapping change, which may affect confidentiality, integrity and availability. Exploitation requires local access with low privileges on a host that uses DPDK with VDUSE. Red Hat rates the security impact as Important.
Red Hat · DPDK
The insights-runtime-extractor runs as a privileged DaemonSet with hostPID and can be tricked into executing a binary from a workload's container. The executed binary runs with root UID in the host user namespace and inherits a working directory that exposes the host CRI-O socket. A local attacker who can schedule a pod on the cluster can therefore compromise the whole node, provided the workload's process name matches a configured fingerprint pattern.
Red Hat · OpenShift Container Platform
A flaw in gnome-remote-desktop as shipped in Red Hat Enterprise Linux lets a remote attacker who can reach an enabled VNC TLS listener drive the event-loop thread into an unbounded busy loop. The thread spins at high CPU without making progress, which causes a denial of service of the remote desktop service. No credentials or user interaction are required, and confidentiality and integrity are not affected, so exposure of the VNC listener is the decisive factor.
Red Hat · gnome-remote-desktop
