CVE-2026-71486

The record

vLLM is an inference and serving engine for large language models. Prior to 0.26.0, the /v1/completions/derender and /v1/chat/completions/derender endpoints accept caller-supplied GenerateResponse objects whose generate_responses, choices, token_ids, prompt_logprobs, logprobs.content, top_logprobs, and routed_experts structures are processed by OnlineDerenderer and tokenizer.decode before max_model_len, max_tokens, max_num_seqs, or response-size limits are enforced, allowing an authenticated API client to consume excessive CPU and memory and produce oversized responses. This issue is fixed in version 0.26.0.

Affected packages

Related vulnerabilities

• CVE-2026-113795 pypdf is a free and open-source pure-python PDF library. CWE-400
• CVE-2026-83400 Vulnerability in the Oracle Java SE, Oracle GraalVM for JDK, Oracle GraalVM Enterprise… CWE-400
• CVE-2026-112400 Pillow is a Python imaging library. CWE-400
• CVE-2026-111766 markdown-it is a Markdown parser. CWE-400
• CVE-2026-111759 markdown-it is a Markdown parser. CWE-400
• CVE-2025-52290 A NULL pointer dereference in the avpriv_h264_has_num_reorder_frames function of Ffmpeg… CWE-400
• CVE-2026-111066 AIOHTTP is an asynchronous HTTP client/server framework for asyncio and Python. CWE-400
• CVE-2026-59951 The MCP Python SDK, called mcp on PyPI, is a Python implementation of the Model Context… CWE-400