CVE-2026-94626

Received Received - Intake

Memory Exhaustion in vLLM via Unbounded tp_size Parameter

Publication date: 2026-09-21

Last updated on: 2026-09-21

Assigner: VulnCheck

Description

vLLM through 0.29.0 fails to validate the tp_size parameter in kv_transfer_params on OpenAI-compatible completion endpoints, allowing attackers to allocate unbounded memory. Attackers can supply arbitrary tp_size values in prefill/decode disaggregated deployments to exhaust memory and trigger kernel OOM-kill of the decode worker process.

CVSS Scores

EPSS Scores

Probability:
Percentile:

Meta Information

Published

2026-09-21

Last Modified

2026-09-21

Generated

2026-09-22

AI Q&A

2026-09-22

EPSS Evaluated

N/A

NVD

CVE-2026-94626

EUVD

EUVD-2026-84220

Affected Vendors & Products

Currently, no data is known.

Helpful Resources

Exploitability

CWE

KEV

CWE ID	Description
CWE-789	The product allocates memory based on an untrusted, large size value, but it does not ensure that the size is within expected limits, allowing arbitrary amounts of memory to be allocated.

Attack-Flow Graph

Executive Summary

vLLM through version 0.29.0 has a flaw where it does not properly validate the tp_size parameter in kv_transfer_params on OpenAI-compatible completion endpoints. This allows attackers to input arbitrary values for tp_size in prefill or decode disaggregated deployments, causing unbounded memory allocation and potentially exhausting system memory.

Detection Guidance

This vulnerability can be detected by monitoring for abnormal memory usage patterns in vLLM deployments, particularly in disaggregated setups. Check for processes consuming excessive memory or frequent OOM-kill events in decode worker processes. Inspect logs for kv_transfer_params with unusually high tp_size values.

Impact Analysis

This vulnerability can lead to denial-of-service conditions by causing memory exhaustion. Attackers could trigger kernel OOM-killer to terminate the decode worker process, disrupting services relying on vLLM for AI model inference or completion tasks.

Compliance Impact

This vulnerability could lead to denial-of-service conditions by exhausting memory, potentially causing service disruptions. Such disruptions may impact data availability, which is a key requirement under GDPR and HIPAA. However, the provided CVE data does not explicitly link this issue to compliance violations.

Mitigation Strategies

Immediately upgrade vLLM to a version that validates the tp_size parameter. Implement input validation to reject non-standard tp_size values. Restrict access to OpenAI-compatible completion endpoints to trusted users only. Monitor memory usage and set alerts for abnormal spikes.

Hi! I’m here to help you understand CVE-2026-94626. Ask me anything about the vulnerability, its impact, or mitigation strategies.

0/70