Summary
CVE-2026-47701 is a high-severity information disclosure vulnerability in the OpenTelemetry Operator's TargetAllocator handling of Prometheus ServiceMonitors. A tenant who can modify a matched ServiceMonitor may cause the Collector to read a sensitive mounted file and send its contents as bearer authorization to a tenant-controlled endpoint.
Affected context
The issue affects OpenTelemetry Operator deployments before 0.152.0 when TargetAllocator Prometheus custom-resource support is enabled. Risk is present where tenant-controlled ServiceMonitors are selected by the Collector, sensitive files are mounted in the Collector pod, and the Collector can connect to a target chosen through the ServiceMonitor configuration.
Exploit preconditions
- Attack surface: Prometheus ServiceMonitor configuration processed by the TargetAllocator
- Required access: Permission to create or update a ServiceMonitor selected by the relevant serviceMonitorSelector and serviceMonitorNamespaceSelector
- Trigger point: Collector scrape configuration generation and subsequent periodic scraping
- Required deployment condition: The Collector pod must contain a service-account token or another sensitive file
- Network condition: The Collector must be able to reach the attacker- or tenant-controlled scrape target
- Why exploitation is possible: The selected bearerTokenFile value is preserved as a credentials-file reference, allowing the Collector to read the file and transmit its contents as an HTTP bearer credential
Why this happens
The underlying bug is an insufficient trust-boundary check when translating ServiceMonitor endpoint configuration into Collector HTTP authentication settings. A tenant-controlled file reference is treated as an acceptable credentials-file path without restricting access to sensitive filesystem locations. This behavior maps to CWE-200: Exposure of Sensitive Information to an Unauthorized Actor, because protected credentials or mounted files can be disclosed through an outbound scrape request.
Potential impact
- A service-account JWT or another mounted secret may be disclosed.
- An attacker could use exposed credentials against Kubernetes APIs within the service account's permissions.
- Sensitive application or infrastructure files may be transmitted outside the intended trust boundary.
- Repeated scrape intervals could cause continued credential exposure.
- The Collector's network position and permissions could increase the consequences of disclosure.
General mitigation guidance
Mitigation generally focuses on reducing trust in tenant-controlled ServiceMonitor fields and preventing monitoring configuration from selecting arbitrary filesystem paths.
Common defensive patterns include:
- Restricting credentials-file references to an explicit, allowlisted set of non-sensitive paths.
- Rejecting ServiceMonitor endpoint configurations that reference protected filesystem locations.
- Applying deny-by-default filesystem access policies for monitoring resources.
- Separating tenant-controlled monitoring configuration from privileged Collector workloads.
- Mounting only the files required by the Collector and avoiding unnecessary service-account token exposure.
- Running Collector service accounts with the minimum Kubernetes permissions required.
- Limiting Collector egress to approved scrape destinations.
- Maintaining deployment configurations that include the relevant filesystem-access control by default.
How teams detect issues like this
Organizations can look for this issue through:
- Static analysis and configuration checks for unvalidated bearerTokenFile or credentials-file propagation.
- Code review signals involving ServiceMonitor fields crossing tenant boundaries into Collector authentication settings.
- Admission policies that reject sensitive filesystem paths in monitoring resources.
- Kubernetes audit review for unexpected ServiceMonitor creation or modification.
- Network monitoring for Collector requests carrying bearer authorization to unapproved destinations.
- File-access and process telemetry showing Collector reads of service-account tokens or unrelated mounted secrets.
- Automated dependency and security review tools that flag missing filesystem-access restrictions or insecure configuration translation.

