CVE-2026-71486

vLLM: Derender endpoints decode caller-supplied GenerateResponse token IDs without output bounds

Published Aug 17, 2026 CVSS 3.1 · 4.3 Medium EPSS < 1% KEV · Not listed

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.

Analysis

Technical impact

The vulnerability allows an authenticated API client to send GenerateResponse objects that are decoded without enforcing maximum response size or token limits. This permits the client to cause the server to consume excessive CPU and memory, resulting in large or unlimited responses, which can lead to denial of service or resource exhaustion.

Affected Systems

The issue affects the vllm-project vllm inference engine before version 0.26.0. All releases prior to 0.26.0 are vulnerable; later releases contain the fix.

Risk and Exploitability

The CVSS score of 4.3 denotes a moderate severity. The EPSS score is not available and the vulnerability is not listed in the CISA KEV catalog, suggesting a lower immediate exploitation risk. However, because the flaw is triggered only by an authenticated user sending a specially crafted request to the derender endpoints, the attack requires valid credentials. The lack of output size checks means an attacker can drain server resources and potentially disrupt service availability within a single authenticated session.

Affected Products

Default status is the baseline for the product, each version can override it (e.g. patched versions marked unaffected).

Vendor Product Default status Versions

vllm-project

vllm

affected

Version	Status	Constraints
`< 0.26.0`	affected	—

Configuration 1 [-]

cpe:2.3:a:vllm:vllm:*:*:*:*:*:*:*:*

No data.

Vendor Product Versions

Vllm-project

Vllm

Version	Status	Scheme	Platform
`[0,0.26.0)`	affected	semver	—

Remediation

No solution or workaround provided in the CVE record.

OpenCVE Recommended Actions

Upgrade the vllm engine to version 0.26.0 or later.
Disable or restrict the /v1/completions/derender and /v1/chat/completions/derender endpoints for unauthenticated or non‑critical traffic.
Monitor CPU and memory usage of the inference service and enforce overall limits at the infrastructure level.

Generated by OpenCVE AI on August 17, 2026 at 21:37 UTC.

Tracking

Advisories

Source	ID	Title
Github GHSA	GHSA-8737-qx52-hjff	vLLM: Derender endpoints decode caller-supplied GenerateResponse token IDs without output bounds

Metrics

No CVSS v4.0

Attack Vector Network

Attack Complexity Low

Privileges Required Low

Scope Unchanged

Confidentiality Impact None

Integrity Impact None

Availability Impact Low

User Interaction None

No CVSS v3.0

No CVSS v2

This CVE is not in the KEV list.

Exploit prediction score: < 1% (0.374% probability within 30 days)

Exploitation none

Automatable no

Technical Impact partial

References

Link	Providers
https://github.com/vllm-project/vllm/commit/8e61b646e2d157f9b93451fa048f9c8530c8a67b
https://github.com/vllm-project/vllm/pull/47260
https://github.com/vllm-project/vllm/releases/tag/v0.26.0
https://github.com/vllm-project/vllm/security/advisories/GHSA-8737-qx52-hjff
https://nvd.nist.gov/vuln/detail/CVE-2026-71486
https://www.cve.org/CVERecord?id=CVE-2026-71486

History

Oct 2, 2026 7:45 PM First appearance in the feed. Affected versions updated. Affected vendors updated View changes

Type	Values Removed	Values Added
First Time appeared		Vllm Vllm vllm
CPEs		cpe:2.3:a:vllm:vllm::::::::
Vendors & Products		Vllm Vllm vllm

Aug 27, 2026 6:00 PM Added references. Scoring metrics updated View changes

Type	Values Removed	Values Added
References		https://nvd.nist.gov/vuln/detail/CVE-2026-71486 https://www.cve.org/CVERecord?id=CVE-2026-71486
Metrics	threat_severity `None`	threat_severity `Moderate`

Aug 18, 2026 1:30 PM Scoring metrics updated View changes

Type	Values Removed	Values Added
Metrics		ssvc `{'options': {'Automatable': 'no', 'Exploitation': 'none', 'Technical Impact': 'partial'}, 'version': '2.0.3'}`

Aug 17, 2026 10:00 PM First appearance in the feed. Affected vendors updated View changes

Type	Values Removed	Values Added
First Time appeared		Vllm-project Vllm-project vllm
Vendors & Products		Vllm-project Vllm-project vllm

Aug 17, 2026 8:15 PM Description updated. Title updated. Weaknesses updated. Added references. Scoring metrics updated View changes

Type	Values Removed	Values Added
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.
Title		vLLM: Derender endpoints decode caller-supplied GenerateResponse token IDs without output bounds
Weaknesses		CWE-400 CWE-770
References		https://github.com/vllm-project/vllm/commit/8e61b646e2d157f9b93451fa048f9c8530c8a67b https://github.com/vllm-project/vllm/pull/47260 https://github.com/vllm-project/vllm/releases/tag/v0.26.0 https://github.com/vllm-project/vllm/security/advisories/GHSA-8737-qx52-hjff
Metrics		cvssV3_1 `{'score': 4.3, 'vector': 'CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:L'}`