CVE-2026-71486
Received Received - Intake

Memory Exhaustion in vLLM API Endpoints

Publication date: 2026-08-17

Last updated on: 2026-08-17

Assigner: GitHub, Inc.

Description
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.

CVSS Scores

EPSS Scores

Probability:
Percentile:

Meta Information

Published
2026-08-17
Last Modified
2026-08-17
Generated
2026-08-18
AI Q&A
2026-08-17
EPSS Evaluated
N/A
NVD
CVE-2026-71486
EUVD
EUVD-2026-60376

Affected Vendors & Products

Showing 1 associated CPE
Vendor Product Version / Range
vllm_project vllm to 0.26.0 (exc)

Helpful Resources

Exploitability

CWE
CWE Icon
KEV
KEV Icon
CWE ID Description
CWE-400 The product does not properly control the allocation and maintenance of a limited resource.
CWE-770 The product allocates a reusable resource or group of resources on behalf of an actor without imposing any intended restrictions on the size or number of resources that can be allocated.

Attack-Flow Graph

AI Quick Actions

Instant insights powered by AI
Executive Summary

This vulnerability in vLLM allows authenticated API clients to send oversized JSON payloads to the /v1/completions/derender and /v1/chat/completions/derender endpoints. The endpoints process these payloads without enforcing resource limits like max_model_len or max_tokens, causing excessive CPU and memory consumption. The issue was fixed in vLLM version 0.26.0 by adding validation to limit payload sizes before processing.

Detection Guidance

Monitor CPU and memory usage on systems running vLLM <= 0.26.0, particularly when derender endpoints are exposed. Check for unusually large JSON payloads sent to /v1/completions/derender or /v1/chat/completions/derender. Inspect logs for excessive token_ids, choices, or generate_responses fields in requests.

Impact Analysis

An attacker could exploit this to degrade service availability by consuming excessive server resources, leading to slower response times or crashes for other users sharing the same vLLM instance. This could disrupt normal operations if the server becomes overloaded.

Compliance Impact

This vulnerability primarily causes resource exhaustion (CPU/memory) due to unbounded payload processing in derender endpoints. While not directly violating GDPR or HIPAA, it could lead to service disruptions that compromise availability requirements under both regulations. GDPR requires ensuring system resilience (Article 32) and HIPAA demands safeguards against unauthorized resource consumption that could impact data processing systems.

Mitigation Strategies

Upgrade vLLM to version 0.26.0 or later to apply resource bounds validation. If upgrading is not immediately possible, restrict access to derender endpoints or disable them entirely. Monitor network traffic for oversized payloads targeting these endpoints.

Chat Assistant

Ask questions about this CVE
Hi! I’m here to help you understand CVE-2026-71486. Ask me anything about the vulnerability, its impact, or mitigation strategies.
0/70

EPSS Chart