Resource exhaustion in Vllm

CVE-2026-71486

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.

Vulnerability class: DoS (Denial of Service)

Published · last modified .

CVSS v3 metric

CVSS v3 base score 4.3 (Medium). Vector: CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:L.

EPSS exploit prediction

EPSS 0.004, 31.8th percentile (scored ); unchanged over 30 days. About EPSS.

Affected products

Weakness classification (CWE)

Public exploits and proof-of-concept code

1 public repository with exploit or proof-of-concept code for this CVE:

  • tmvictorpeters/jbo4rgl (last pushed 2026-10-05): r30xlqo6从一个辅助 API 看资源边界:CVE-2026-71486(vLLM)w680rjb2afux

References