Veille CVE

Mercredi 7 octobre 2026

5 CVE

Critique 0 Élevée 2 Moyenne 3 Faible 0 Non évaluée 0 Exploitées (KEV) Tri : plus récentes « Coder coder » ✕
  1. CVE-2026-19029 6.8 Moyenne

    The HDF Group HDF5

    A heap-based buffer over-read in H5Z__filter_scaleoffset() in src/H5Zscaleoffset.c in HDF5 through 2.2.0 lets an attacker cause a denial of service (application crash) with a crafted HDF5 file. When the stored minimum bits equal the full precision of the datatype, the decoder copies d_nelmts * size bytes from the compressed chunk without checking that the chunk holds that many bytes. Both values come from attacker-controlled scale-offset filter parameters in the dataset's filter pipeline message.

  2. CVE-2026-106447 8.7 Élevée

    stablelib

    StableLib is a stable library of useful TypeScript and JavaScript code. Prior to 2.0.4, the @stablelib/cbor decoder recursively processes nested CBOR arrays, maps, and tags through _decodeValue() without enforcing a maximum nesting depth. A sufficiently deep structure exhausts the JavaScript call stack, causing a decoding exception and potentially terminating an uncaught request worker or process. This issue is fixed in version 2.0.4.

  3. CVE-2026-104677 7.2 Élevée

    Unknown WP Coder

    The WP Coder WordPress plugin before 4.5.2 does not restrict access to its PHP code-execution feature to administrators, gating it on a content capability that the Editor role holds by default, which allows Editor-level users to save and execute arbitrary PHP code on the server and fully compromise the site.

  4. CVE-2026-57173 6.5 Moyenne

    vllm-project vllm · Vllm vllm

    vLLM is an inference and serving engine for large language models. Prior to 0.24.0, the input_audio handling path for /v1/chat/completions calls AudioMediaIO.load_bytes or AudioMediaIO.load_file without passing VLLM_MAX_AUDIO_DECODE_DURATION_S to the shared audio decoder. An unauthenticated client can therefore submit a small compressed audio input that expands into a very large float32 PCM allocation, bypassing the duration guard already used by /v1/audio/transcriptions and causing an out-of-memory worker crash. Inline data URLs reach this path without being bounded by VLLM_AUDIO_FETCH_TIMEOUT. The issue affects deployments serving an audio-capable model, and authentication changes only the deployment-specific reachability. This issue is fixed in version 0.24.0.

  5. CVE-2026-69147 6.5 Moyenne

    vllm-project vllm · Vllm vllm

    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.