CVE-2026-44432

urllib3 is an HTTP client library for Python. urllib3's streaming API is designed for the efficient handling of large HTTP responses by reading the content in chunks, rather than loading the entire response body into memory at once. urllib3 can perform decoding or decompression based on the HTTP `Content-Encoding` header (e.g., `gzip`, `deflate`, `br`, or `zstd`). When using the streaming API, the library decompresses only the necessary bytes, enabling partial content consumption. Starting in version 1.22 and prior to version 2.6.3, for HTTP redirect responses, the library would read the entire response body to drain the connection and decompress the content unnecessarily. This decompression occurred even before any read methods were called, and configured read limits did not restrict the amount of decompressed data. As a result, there was no safeguard against decompression bombs. A malicious server could exploit this to trigger excessive resource consumption on the client. Applications and libraries are affected when they stream content from untrusted sources by setting `preload_content=False` when they do not disable redirects. Users should upgrade to at least urllib3 v2.6.3, in which the library does not decode content of redirect responses when `preload_content=False`. If upgrading is not immediately possible, disable redirects by setting `redirect=False` for requests to untrusted source.

The undici WebSocket client is vulnerable to a denial-of-service attack via unbounded memory consumption during permessage-deflate decompression. When a WebSocket connection negotiates the permessage-deflate extension, the client decompresses incoming compressed frames without enforcing any limit on the decompressed data size. A malicious WebSocket server can send a small compressed frame (a "decompression bomb") that expands to an extremely large size in memory, causing the Node.js process to exhaust available memory and crash or become unresponsive. The vulnerability exists in the PerMessageDeflate.decompress() method, which accumulates all decompressed chunks in memory and concatenates them into a single Buffer without checking whether the total size exceeds a safe threshold.

NVIDIA Triton Inference Server for Linux contains a vulnerability where an attacker can cause improper handling of highly compressed data. A successful exploit of this vulnerability might lead to denial of service.

Envoy is an open source edge and service proxy designed for cloud-native applications. From 1.23.0 until 1.35.11, 1.36.7, 1.37.3, and 1.38.1, a vulnerability has been identified in Envoy's zstd decompressor implementation (ZstdDecompressorImpl). When zstd decompression is enabled, processing a specially crafted, highly compressed zstd payload can lead to massive memory allocation. An attacker can exploit this to cause severe memory exhaustion, potentially resulting in an Out-Of-Memory (OOM) kill and Denial of Service (DoS) for the Envoy proxy. This vulnerability is fixed in 1.35.11, 1.36.7, 1.37.3, and 1.38.1.

AIOHTTP is an asynchronous HTTP client/server framework for asyncio and Python. Prior to 3.14.1, during cleanup it is possible for a compressed request body to be decompressed into memory in one chunk. An attacker may be able to send a compressed payload in specific situations that could be decompressed into memory, potentially leading to DoS (a zip bomb edge case). This vulnerability is fixed in 3.14.1.

vLLM is an inference and serving engine for large language models (LLMs). Prior to 0.23.1rc0, vLLM's /v1/audio/transcriptions endpoint limits compressed upload size but not decoded PCM output. A 25MB OPUS file expands to ~14.9GB of float32 PCM at decode time. This vulnerability is fixed in 0.23.1rc0.