CVE-2026-115911

The record

vLLM is an inference and serving engine for large language models. From 0.19.0 until 0.32.0, authenticated callers can supply a request chat template when no server-side --chat-template is configured, or a malicious model repository can provide a template that is compiled and rendered by tokenizer.apply_chat_template() without a wall-clock timeout, output-size limit, or iteration bound. Rendering runs in a ThreadPoolExecutor with one worker by default, so a nonterminating render can block /v1/responses and other chat-template users after the request ends, while an output-expanding render can exhaust server memory. Harmony models are not affected because they do not use this Jinja path. This issue is fixed in version 0.32.0.

Affected packages

Related vulnerabilities

• CVE-2026-63341 Aedes is an MQTT broker that can run on any stream server. CWE-400
• CVE-2026-62300 Filecoin Pin is an IPFS Pinning Service API implementation that pins to Filecoin's PDP… CWE-400
• CVE-2026-62254 LibreDWG is a free C library to read and write DWG files. CWE-400
• CVE-2026-62250 LibreDWG is a free C library to read and write DWG files. CWE-400
• CVE-2026-62303 Pistache is a high-performance REST toolkit written in C++. CWE-400
• CVE-2026-62160 Daptin is a GraphQL/JSON-API headless CMS. CWE-400
• CVE-2026-61848 Crass is a Ruby CSS parser that's fully compliant with the CSS Syntax Level 3… CWE-400
• CVE-2026-61854 VoiceSentinel is a real-time voice moderation system for Minecraft servers. CWE-400