OpenTelemetry.Sampler.AWS EUVD-2026-25271

Exploitation requires one of two conditions: (1) the attacker controls the value passed to AWSXRayRemoteSamplerBuilder.SetEndpoint, which implies write access to the application's configuration at deployment time (administrative or infrastructure-level access), or (2) the attacker holds a network MitM position on the path to the configured sampling endpoint. For the default configuration (http://localhost:2000), a MitM attack requires local code execution or a compromised process on the same host, substantially raising the bar. In containerized environments using service meshes or shared network namespaces, the MitM requirement may be achievable via lateral movement. For OpenTelemetry.Resources.AWS, the metadata endpoints (http://169.254.169.254 for EC2, link-local ECS metadata, https://kubernetes.default.svc for EKS) are only reachable from within the instance or pod, so exploitation similarly requires prior foothold in the execution environment. Exploitation is not automatable at scale per SSVC, and no authentication bypass is involved - the attacker manipulates the response rather than the request.

The CVSS 3.1 score of 5.9 (AV:N/AC:H/PR:N/UI:N/S:U/C:N/I:N/A:H) reflects high availability impact but elevated attack complexity, which accurately captures the prerequisite of either endpoint control or a network MitM position. EPSS at 0.01% (2nd percentile) independently corroborates the low exploitation likelihood. SSVC classifies exploitation as none and the attack as not automatable, further lowering operational urgency. The default endpoint of http://localhost:2000 substantially limits MitM feasibility in standard deployments, as intercepting loopback traffic requires local process access or a compromised sidecar. Risk increases meaningfully in container orchestration environments (Kubernetes, ECS) where the metadata endpoints are link-local or service-mesh addresses and where lateral movement is more plausible. There is no confidentiality or integrity impact - the sole outcome is process destabilization or termination. No conflict between signals is present; all sources align on a medium-severity, low-probability DoS.

An attacker with a MitM position on the network path between the .NET application and the X-Ray daemon (for example, through ARP spoofing on an internal segment or by compromising a sidecar proxy in a Kubernetes pod) intercepts the periodic HTTP GET request the sampler makes to http://localhost:2000/GetSamplingRules. The attacker responds with a multi-gigabyte body; the unpatched ReadAsStringAsync() call buffers the entire payload on the .NET heap, triggering garbage-collection stalls and ultimately an OutOfMemoryException that terminates the instrumented process. No public proof-of-concept exploit code has been identified at time of analysis.