CVE-2026-114055: vLLM unvalidated mm_processor_kwargs

Q: Is CVE-2026-114055 actively exploited?

No confirmed active exploitation of CVE-2026-114055 has been reported, but organizations should still patch proactively.

Q: How to fix CVE-2026-114055?

1) Upgrade vLLM to 0.31.0 or later. 2) Until patched, place an API gateway or reverse proxy in front of vLLM that rejects or strips mm_processor_kwargs and enforces maximum request body, image dimension, and frame count limits. 3) Require authentication and per-client rate limits on the inference endpoint, and avoid exposing it directly to the internet. 4) Set container and host memory limits with automatic restart and health checks, and isolate tenants or workloads where possible. 5) Detect by alerting on sudden memory spikes, OOM kills, repeated server restarts, and requests containing unusually large mm_processor_kwargs values. 6) Review GHSA-gx7p-2j49-hfq4 for the full list of affected processors.

Q: What systems are affected by CVE-2026-114055?

This vulnerability affects the following AI/ML architecture patterns: model serving, multimodal inference, LLM inference APIs, multi-tenant GPU inference clusters.

Q: What is the CVSS score for CVE-2026-114055?

CVE-2026-114055 has a CVSS v3.1 base score of 7.5 (HIGH). The EPSS exploitation probability is 0.42%.

CISO Take

vLLM versions 0.7.3 up to (but not including) 0.31.0 let any remote client pass per-request multimodal processor parameters (image size, patch, crop, or frame counts) that are not validated, so a single crafted request can exhaust memory and crash the shared inference server. The CVSS 7.5 rating reflects network reachability, no authentication, and low complexity, which means anyone who can reach the API endpoint can take down inference for every tenant. No active exploitation is recorded: it is not in CISA KEV, no public exploit or Nuclei template is listed, and EPSS and SSVC data are not yet available, so urgency is driven by how trivial the attack is rather than by observed abuse. Upgrade to vLLM 0.31.0 or later, and until then block or strip client-supplied mm_processor_kwargs at a gateway and set container memory limits with automatic restart.

Sources:NVDGitHub AdvisoryVulnCheckATLAS

What is the risk?

High for availability, none for confidentiality or integrity. Exploitation needs only network access to the OpenAI-compatible API and one request with oversized processor parameters, with no privileges or user interaction. Exposure is highest for internet-facing or multi-tenant vLLM deployments serving multimodal models (Isaac, InternVL, H2OVL and similar). Likelihood is moderate to high once details are public, since the technique is trivial, though no exploitation or public PoC is known today. Data gaps: EPSS, downstream dependent counts, and package scorecard are unavailable.

How does the attack unfold?

Initial Access

Attacker reaches the vLLM inference API endpoint over the network without authentication.

AML.T0040

Exploitation

Attacker submits a request with oversized mm_processor_kwargs values for image size, patches, crops, or frames.

AML.T0034.001

Impact

The multimodal processor exhausts memory and the shared server terminates, denying service to all users.

AML.T0029

Initial Access

Attacker reaches the vLLM inference API endpoint over the network without authentication.

AML.T0040

Exploitation

Attacker submits a request with oversized mm_processor_kwargs values for image size, patches, crops, or frames.

AML.T0034.001

Impact

The multimodal processor exhausts memory and the shared server terminates, denying service to all users.

AML.T0029

What systems are affected?

Package	Ecosystem	Vulnerable Range	Patched
vLLM	pip	>= 0.7.3, < 0.31.0	`0.31.0`
95.4K94 dependentsPushed 6d ago27% patched~37d to patchFull package profile →

Do you use vLLM? You're affected.

How severe is it?

CVSS 3.1

7.5 / 10

EPSS

0.4%

chance of exploitation in 30 days

Higher than 36% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

No known exploitation

Sophistication

Trivial

What is the attack surface?

AVNetwork

ACLow

PRNone

UINone

SUnchanged

CNone

INone

AHigh

What should I do?

1 step

1) Upgrade vLLM to 0.31.0 or later. 2) Until patched, place an API gateway or reverse proxy in front of vLLM that rejects or strips mm_processor_kwargs and enforces maximum request body, image dimension, and frame count limits. 3) Require authentication and per-client rate limits on the inference endpoint, and avoid exposing it directly to the internet. 4) Set container and host memory limits with automatic restart and health checks, and isolate tenants or workloads where possible. 5) Detect by alerting on sudden memory spikes, OOM kills, repeated server restarts, and requests containing unusually large mm_processor_kwargs values. 6) Review GHSA-gx7p-2j49-hfq4 for the full list of affected processors.

What does CISA's SSVC say?

DecisionTrack

Exploitationnone

AutomatableYes

Technical Impactpartial

Source: CISA Vulnrichment (SSVC v2.0). Decision based on the CISA Coordinator decision tree.

How is it classified?

DoSInferenceAPIAML.T0029- Denial of AI Service AML.T0034.001- Resource-Intensive Queries AML.T0040- AI Model Inference API Access

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Art. 15-Accuracy, robustness and cybersecurity

ISO 42001

A.6.2.6-AI system operation and monitoring

NIST AI RMF

MANAGE 2.3-Mechanisms to respond to and recover from unknown risks

OWASP LLM Top 10

LLM10-Unbounded Consumption

Frequently Asked Questions

What is CVE-2026-114055?

vLLM versions 0.7.3 up to (but not including) 0.31.0 let any remote client pass per-request multimodal processor parameters (image size, patch, crop, or frame counts) that are not validated, so a single crafted request can exhaust memory and crash the shared inference server. The CVSS 7.5 rating reflects network reachability, no authentication, and low complexity, which means anyone who can reach the API endpoint can take down inference for every tenant. No active exploitation is recorded: it is not in CISA KEV, no public exploit or Nuclei template is listed, and EPSS and SSVC data are not yet available, so urgency is driven by how trivial the attack is rather than by observed abuse. Upgrade to vLLM 0.31.0 or later, and until then block or strip client-supplied mm_processor_kwargs at a gateway and set container memory limits with automatic restart.

Is CVE-2026-114055 actively exploited?

No confirmed active exploitation of CVE-2026-114055 has been reported, but organizations should still patch proactively.

How to fix CVE-2026-114055?

1) Upgrade vLLM to 0.31.0 or later. 2) Until patched, place an API gateway or reverse proxy in front of vLLM that rejects or strips mm_processor_kwargs and enforces maximum request body, image dimension, and frame count limits. 3) Require authentication and per-client rate limits on the inference endpoint, and avoid exposing it directly to the internet. 4) Set container and host memory limits with automatic restart and health checks, and isolate tenants or workloads where possible. 5) Detect by alerting on sudden memory spikes, OOM kills, repeated server restarts, and requests containing unusually large mm_processor_kwargs values. 6) Review GHSA-gx7p-2j49-hfq4 for the full list of affected processors.

What systems are affected by CVE-2026-114055?

This vulnerability affects the following AI/ML architecture patterns: model serving, multimodal inference, LLM inference APIs, multi-tenant GPU inference clusters.

What is the CVSS score for CVE-2026-114055?

CVE-2026-114055 has a CVSS v3.1 base score of 7.5 (HIGH). The EPSS exploitation probability is 0.42%.

What is the AI security impact?

Affected AI Architectures

model servingmultimodal inferenceLLM inference APIsmulti-tenant GPU inference clusters

MITRE ATLAS Techniques

AML.T0029Denial of AI Service

AML.T0034.001Resource-Intensive Queries

AML.T0040AI Model Inference API Access

Compliance Controls Affected

EU AI Act: Art. 15

ISO 42001: A.6.2.6

NIST AI RMF: MANAGE 2.3

OWASP LLM Top 10: LLM10

What are the technical details?

Original Advisory

vLLM from 0.7.3 before 0.31.0 contains an uncontrolled resource consumption vulnerability that allows remote clients to control multimodal processor parameters by supplying per-request mm_processor_kwargs. Attackers can set large image size, patch count, crop count, or frame count values in Isaac, InternVL, H2OVL, and other processors to exhaust memory and terminate the shared server.

Exploitation Scenario

An attacker finds an exposed or internally reachable vLLM OpenAI-compatible endpoint serving a multimodal model such as InternVL. They send a chat completion request with mm_processor_kwargs setting extremely large values for patch count, crop count, or frame count alongside a small image. The processor allocates excessive memory, the server process is OOM-killed, and all applications depending on it lose inference until it is restarted. The attacker repeats the request after each restart to sustain the outage.

Weaknesses (CWE)

CWE-400Uncontrolled Resource ConsumptionPrimary CWE-400Uncontrolled Resource ConsumptionPrimary CWE-400Uncontrolled Resource Consumption

CWE-400 — Uncontrolled Resource Consumption: The product does not properly control the allocation and maintenance of a limited resource.

[Architecture and Design] Design throttling mechanisms into the system architecture. The best protection is to limit the amount of resources that an unauthorized user can cause to be expended. A strong authentication and access control model will help prevent such attacks from occurring in the first place. The login application should be protected against DoS attacks as much as possible. Limiting the database access, perhaps by caching result sets, can help minimize the resources expended. To further limit the potential for a DoS attack, consider tracking the rate of requests received from users and blocking requests that exceed a defined rate threshold.
[Architecture and Design] Mitigation of resource exhaustion attacks requires that the target system either: The first of these solutions is an issue in itself though, since it may allow attackers to prevent the use of the system by a particular valid user. If the attacker impersonates the valid user, they may be able to prevent the user from accessing the server in question. The second solution is simply difficult to effectively institute -- and even when properly done, it does not provide a full solution. It simply makes the attack require more resources on the part of the attacker. recognizes the attack and denies that user further access for a given amount of time, or uniformly throttles all requests in order to make it more difficult to consume resources more quickly than they can again be freed.

Source: MITRE CWE corpus.