CVE-2026-76646
A remote attacker could cause excessive resource consumption by supplying specially crafted request parameters, potentially resulting in a d
CVSS
—
No CVSS
EPSS
—
KEV
—
Exploit Today
—
0-100
Published: Sep 16, 2026 · Last modified: Sep 16, 2026 · CWE-400
Not enough EPSS history yet.
A remote attacker could cause excessive resource consumption by supplying specially crafted request parameters, potentially resulting in a denial of service condition. Older unsupported versions may also be affected. Users are recommended to upgrade to versions 2.3.12, 2.3-next-M9, 3.0.4, 4.0.4, or 4.1.4, which fix this issue.
CVECVSSEPSSKEVRExploitTitleMod.
CVE-2026-925967.5 HIG—
———Nodemailer before 9.1.0 contains a quadratic time complexity vulnerability in the addressparser component that allows remote attackers to cause denial of service by supplying a crafted comma-separated address list. Attackers can send a single email with a large number of addresses to block the Node.js event loop for extended periods, consuming 100% CPU and freezing the process.14hCVE-2026-81872——
———OpenTelemetry-Go is the Go implementation of OpenTelemetry. Prior to version 0.21.0, the go.opentelemetry.io/otel/sdk/log BatchingProcessor can enter a tight CPU loop when attacker-driven log emission fills its asynchronous export buffer while the exporter is backpressured. NewBatchingProcessor wraps the exporter with newBufferExporter(exporter, 1), and the poll loop calls queue.TryDequeue and bufferExporter.EnqueueExport before immediately signaling pollTrigger whenever the queue remains at or above batchSize. Because a failed nonblocking EnqueueExport leaves the queue length unchanged, the processor repeatedly retries without waiting for its ticker, exhausting CPU and degrading or denying service in the embedding process. This issue is fixed in version 0.21.0.15hCVE-2026-81869——
———OpenTelemetry-Go is the Go implementation of OpenTelemetry. From version 1.10.0 until 1.33.0, the sdk/trace/span.go attribute truncation path can fail to enforce AttributeValueLengthLimit for string and string-slice attributes containing the valid Unicode replacement character U+FFFD. safeTruncateValidUTF8 treats the valid replacement rune as invalid UTF-8 and returns the original input, while strings.ToValidUTF8 leaves that valid rune unchanged, so a second safeTruncate attempt can also return the oversized value. An attacker who controls span attribute content can retain values longer than the configured limit, increasing per-span memory use and weakening denial-of-service protection in the instrumented process. This issue is fixed in version 1.33.0.15hCVE-2026-818767.5 HIG—
———HAPI FHIR is a complete implementation of the HL7 FHIR standard for healthcare interoperability in Java. Prior to version 6.9.12, SHCParser in org.hl7.fhir.r5/src/main/java/org/hl7/fhir/r5/elementmodel/SHCParser.java can enter an infinite loop while processing attacker-controlled Smart Health Card JWT content whose header contains zip: "DEF" and whose raw-DEFLATE payload is empty or truncated. SHCParser.decodeJWT() reaches SHCParser.inflate(), where Inflater.inflate() can return zero while Inflater.finished() remains false and Inflater.needsInput() is true. The loop also lacks an Inflater.needsDictionary() termination check, SHCParser.decompress() contains the same zero-progress pattern, and ResourceChecker.java can reach SHC parsing during file-format detection. A malformed validation request can pin a JVM worker thread indefinitely, and concurrent requests can exhaust all validation workers. This issue is fixed in version 6.9.12.17hCVE-2026-818757.5 HIG—
———HAPI FHIR is a complete implementation of the HL7 FHIR standard for healthcare interoperability in Java. Prior to version 6.9.12, SHCParser in org.hl7.fhir.r5/src/main/java/org/hl7/fhir/r5/elementmodel/SHCParser.java can consume attacker-controlled Smart Health Card JWT content whose header contains zip: "DEF" and whose small raw-DEFLATE payload expands to a very large value. SHCParser.decodeJWT() passes the decoded payload to SHCParser.inflate(), which accumulates all decompressed bytes in a ByteArrayOutputStream without an output-size limit before JSON parsing, and SHCParser.decompress() contains the same unbounded pattern. An application or validator service that accepts attacker-supplied SHC content can therefore suffer excessive heap allocation, severe garbage-collection pressure, request failure, process instability, or process termination. This issue is fixed in version 6.9.12.17hCVE-2026-691476.5 MED—
———vLLM is an inference and serving engine for large language models. Prior to 0.28.0, request bodies for Chat Completions and Responses can set media_io_kwargs.video.video_backend to pynvvideocodec, and MediaConnector.fetch_video forwards that choice to VideoMediaIO even when startup configuration selected a software decoder. The engine's _reserve_mm_ipc_gpu_memory logic budgets decoder memory only from static configuration, so the request-selected VIDEO_LOADER_REGISTRY backend can create a CUDA context, decoder surfaces, and decoded-frame allocations that were not removed from the engine's KV-cache budget. An attacker able to submit video requests to a video-capable GPU deployment with PyNvVideoCodec installed can exhaust shared GPU memory, causing request failures, worker crashes, or denial of service. The first release containing the fix is version 0.28.0.17h