CVE-2026-105759
vLLM is an inference and serving engine for large language models. Prior to 0.30.0, the Rust frontend's track_http_metrics middleware record
CVSS
5.9
Medio
EPSS
0.4%
p29
KEV
—
Exploit Today
9
0-100
Publicado: 5 oct 2026 · Última mod.: 16 oct 2026 · CWE-400
0.3%EPSS · 30 días0.4%
2026-10-062026-10-21
vLLM is an inference and serving engine for large language models. Prior to 0.30.0, the Rust frontend's track_http_metrics middleware records the raw HTTP method token as a Prometheus label for requests reaching registered routes. An unauthenticated attacker can send unique arbitrary method tokens to unguarded routes such as /tokenize, causing Prometheus's Family::get_or_create function to permanently create counter and histogram label sets. Those label sets increase process memory usage and enlarge the /metrics response until the service or monitoring path is exhausted. This issue is fixed in version 0.30.0.
- github.comhttps://github.com/vllm-project/vllm/commit/3735c2d5f5248259482b9045c34fb7a8a3892352
- github.comhttps://github.com/vllm-project/vllm/pull/56058
- github.comhttps://github.com/vllm-project/vllm/releases/tag/v0.30.0
- github.comhttps://github.com/vllm-project/vllm/security/advisories/GHSA-5fj9-pfhr-6j48
- github.comhttps://github.com/vllm-project/vllm/security/advisories/GHSA-5fj9-pfhr-6j48
CVECVSSEPSSKEVRExplotTítuloVis.
CVE-2026-762907.5 ALT—
———In Splunk Enterprise versions below 10.4.4, 10.2.8, and 10.0.11, an unauthenticated user could send repeated requests to the Edge Processor package download endpoint in the Splunk Pipeline Builders app. The requests could cause the Splunk daemon to consume excessive system resources and potentially stop responding.
The vulnerability is possible because the endpoint accepts package download requests without authentication and reads and base64-encodes the complete package in memory for each request.
After upgrading, prepare all existing Edge Processor instances, turn on authentication for Edge Processor package downloads, and restart Splunk Enterprise. For more information, see Enable authentication for Edge Processor package downloads (https://help.splunk.com/en/data-management/process-data-at-the-edge/use-edge-processors-for-splunk-enterprise/10.4/administer-edge-processors/enable-authentication-for-edge-processor-package-downloads).13hCVE-2026-762887.5 ALT—
———In Splunk Enterprise versions below 10.4.4, 10.2.8, 10.0.11, and 9.4.16, an unauthenticated user could send a single crafted request to the Splunk Web JavaScript logger endpoint and cause Splunk Web to stop responding to requests. Splunk Web remains unavailable until Splunk Enterprise is restarted.
The vulnerability is possible because Splunk Web does not limit request-header parsing before processing the request.13hCVE-2026-96640——
———Pillow is a Python imaging library. From 5.1.0 until 13.0.0, PdfParser.linearize_page_tree() in src/PIL/PdfParser.py treats a PDF /Pages graph as a tree without tracking duplicate indirect references. When Image.save(..., format="PDF", append=True) processes an attacker-controlled existing PDF whose /Kids arrays repeatedly reference the same child /Pages node, Pillow expands the shared subtree as distinct pages. A compact malformed PDF can therefore create exponentially many page references and consume disproportionate CPU before the append completes. This issue is fixed in version 13.0.0.1dCVE-2026-966386.5 MED—
———Pillow is a Python imaging library. From 9.1.0 until 13.0.0, BmpRleDecoder.decode() in src/PIL/BmpImagePlugin.py permits crafted RLE BMP delta escapes and absolute-mode runs to append beyond the remaining dest_length without a per-branch cap. A delta can add roughly 255 image rows in one iteration, so a kilobyte-sized file with a large but permitted width can expand the data bytearray to hundreds of megabytes or more, causing memory-exhaustion denial of service. This issue is fixed in version 13.0.0.1dCVE-2026-96637——
———Pillow is a Python imaging library. Prior to 13.0.0, SpiderImageFile.seek() in src/PIL/SpiderImagePlugin.py reparses a later SPIDER stack frame and replaces self._size without rerunning Image._decompression_bomb_check(). A tiny file can use a safe first frame to pass Image.open() and then declare very large dimensions in a later mode F frame. Loading that frame allocates approximately four bytes per attacker-declared pixel and can consume multiple gigabytes of memory. This issue is fixed in version 13.0.0.1dCVE-2026-966366.5 MED—
———Pillow is a Python imaging library. Prior to 13.0.0, MpoImageFile.seek() and DcxImageFile.seek() reparse later MPO or DCX frames and replace self._size without rerunning Image._decompression_bomb_check(). A crafted multi-frame image can use a tiny first frame to pass Image.open() and then declare an over-limit later frame. Normal frame iteration through seek(), load(), or ImageSequence.Iterator can decode that frame and commit hundreds of megabytes or more, causing memory-exhaustion denial of service. This issue is fixed in version 13.0.0.1d