cve.bz

CVE-2026-116177 vllm vllm: Uncontrolled Resource Consumption

8.7 highEPSS 0.3%Fix available

High (8.7) in vllm vllm. EPSS 0.3% estimated probability of exploitation activity in the next 30 days. Fixed in vllm 0.25.0.

On this page
vLLM from 0.10.2 before 0.25.0 contains an uncontrolled resource consumption vulnerability that allows unauthenticated attackers to exhaust memory because the Nemotron-VL processors globally set Image.MAX_IMAGE_PIXELS to None. Attackers can submit a pathological decompression-bomb image to /v1/chat/completions to OOM-kill the worker and deny service to co-tenants.

Severity

CVSS 4.0

8.7 high
CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:N/VI:N/VA:H/SC:N/SI:N/SA:N

CVSS 3.1

7.5 high
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H
EPSS
0.3% (unchanged from 0.3% a week ago) estimated probability of exploitation activity in the next 30 days; 22nd percentile. Scored 2026-11-17. A forecast for this CVE across the internet, not for your own systems.
CISA's assessment (SSVC)
Exploitation: none · Automatable: yes · Technical impact: partial, assessed 2026-10-29. From CISA's Vulnrichment in the CVE List: whether it is exploited, whether exploiting it can be automated, and how much control it gives.
CWE
CWE-400 Uncontrolled Resource Consumption
Published
Modified
NVD analysis
Analyzed
Record from
the NVD, synced

Fix and patched versions

Ask an AI about this CVE

Show the promptExplain CVE-2026-116177 for an engineer deciding whether to act. Use https://cve.bz/api/public/cves/CVE-2026-116177?via=ask-ai-copy and https://cve.bz/cve/CVE-2026-116177?via=ask-ai-copy as the source of truth: what is affected and which versions fix it, whether it is in CISA KEV, public exploits, and its EPSS (an internet-wide 30-day estimate of exploitation activity, not a forecast for my systems). If they give no fixed versions, use the CVE record https://cveawg.mitre.org/api/cve/CVE-2026-116177 or the vendor advisory, not third-party mirrors. Tell me how to check whether I am affected and how to verify a fix. Cite the sources and say where they are uncertain.

Opens a third-party assistant; its answer is its own, not cve.bz's. The prompt holds only this CVE's id and public links.

Affected

Versions

From the CNA's record in the CVE List.

ProductAffected
vllm-project vllmfrom 0.10.2 before 0.25.0

Packages (OSV)

AdvisoryEcosystemPackageAffectedFixed in
PYSEC-2026-4352PyPIvllm>=0.10.2 <0.25.00.25.0

Red Hat

Red Hat rates this Important.

Red Hat's statement

This vulnerability is rated as having Important severity because unauthenticated remote attackers can trigger an out-of-memory crash of inference worker processes by transmitting malicious image payloads across network endpoints without requiring elevated privileges. In Red Hat OpenShift AI and Red Hat Enterprise Linux AI deployments serving vision-language workloads, exhausting host or container memory disrupts service availability for all tenant workloads sharing the affected inference server. However, the flaw is confined to resource exhaustion and denial of service, without enabling arbitrary code execution or data leakage.

Red Hat's mitigation

Restrict network exposure of the vLLM inference API endpoints so that only authenticated, trusted clients can submit requests, or avoid deploying multimodal Nemotron-VL models that accept image payloads.

For standalone deployments on Red Hat Enterprise Linux AI, configure firewalld to limit access to the inference service port (for example, default port 8000) to authorized source addresses:

# firewall-cmd --zone=trusted --add-source=<TRUSTED_IP_OR_SUBNET> --permanent

# firewall-cmd --zone=public --remove-port=8000/tcp --permanent

# firewall-cmd --reload

Warning: Reloading the firewall configuration will momentarily interrupt active network connections.

In containerized environments such as Red Hat OpenShift AI (RHOAI), enforce OpenShift NetworkPolicies to restrict ingress traffic to model-serving pods from trusted namespaces only, and enforce request size limits and client authentication at an upstream ingress controller or API gateway.

Caveats: Restricting endpoint reachability blocks unauthenticated clients from invoking the API. Legitimate clients must connect from trusted networks or authenticate via an upstream proxy.

--- CME Mitigation Guidance (2026-11-02) ---

Disclaimer: The following mitigation solutions are provided as a courtesy, pending the application of an official patch. This content was generated utilizing AI tools in conjunction with data from https://cmetaxonomy.org. Customers are advised to independently evaluate and test these temporary mitigations within their own infrastructure.

The highest-value action is to enforce bounds on decoded image pixel counts and preserve decompression-bomb safeguards in vLLM Nemotron-VL image processing: upgrade to vLLM 0.25.0 (or the corresponding fixed Red Hat builds), which remove module-level Image.MAX_IMAGE_PIXELS assignments and enforce a configurable decoded pixel-count limit via the VLLM_MAX_IMAGE_PIXELS environment variable, closing pathological decompression-bomb image upload via /v1/chat/completions. Until patched, this is a concern on deployments serving multimodal workloads using Nemotron-VL processors, so restrict network access to the /v1/chat/completions endpoint to trusted, authenticated clients, avoid deploying multimodal Nemotron-VL models, and enforce request size limits at upstream ingress gateways. As a compensating measure, set address-space and memory limits on the inference process via ulimit or prlimit to prevent decompression from exhausting the entire node, configure systemd watchdog monitoring and automatic restart directives to recover worker services after an out-of-memory crash, apply cgroups v2 memory limits to constrain worker memory consumption and prevent host resource exhaustion, and monitor cgroup memory pressure metrics using memory.events and PSI to detect anomalous pressure spikes and OOM kill events. These controls limit blast radius and contain exploitation, but they do not fix the underlying decompression-bomb safeguard bypass defect — apply the upstream patch to fully remediate.

Highest-impact CME: CME-1311 Input Size and Quantity Bounds Enforcement (high)

ProductPackageStateAdvisory
Red Hat AI Inference Server 3.5rhaii/vllm-cpu-rhel9:1793444554Fixed 2026-11-03RHSA-2026:94702
Red Hat AI Inference Server 3.5rhaii/vllm-rocm-rhel9:1793444456Fixed 2026-11-03RHSA-2026:94709
Red Hat AI Inference Server 3.5rhaii/vllm-spyre-rhel9:1793444409Fixed 2026-11-03RHSA-2026:94717
Red Hat AI Inference Server 3.5rhaii/vllm-gaudi-rhel9:1793444510Fixed 2026-11-03RHSA-2026:94721
Red Hat AI Inference Server 3.5rhaii/vllm-cuda-rhel9:1793444600Fixed 2026-11-03RHSA-2026:94738
Red Hat AI Inference Serverrhaii-fast/vllm-cpu-rhel9Affected
Red Hat AI Inference Serverrhaii-fast/vllm-cuda-rhel9Affected
Red Hat AI Inference Serverrhaii-fast/vllm-gaudi-rhel9Affected
Red Hat AI Inference Serverrhaii-fast/vllm-neuron-rhel9Will not fix
Red Hat AI Inference Serverrhaii-fast/vllm-omni-cuda-rhel9Will not fix
Red Hat AI Inference Serverrhaii-fast/vllm-rocm-rhel9Affected
Red Hat AI Inference Serverrhaii-fast/vllm-spyre-rhel9Affected
Red Hat AI Inference Serverrhaii-fast/vllm-tpu-rhel9Will not fix
Red Hat AI Inference Serverrhaii/vllm-neuron-rhel9Will not fix
Red Hat AI Inference Serverrhaii/vllm-tpu-rhel9Will not fix
Red Hat Enterprise Linux AI (RHEL AI) 3rhelai3/bootc-aws-cuda-rhel9Affected
Red Hat Enterprise Linux AI (RHEL AI) 3rhelai3/bootc-azure-cuda-rhel9Affected
Red Hat Enterprise Linux AI (RHEL AI) 3rhelai3/bootc-azure-rocm-rhel9Affected
Red Hat Enterprise Linux AI (RHEL AI) 3rhelai3/bootc-cuda-rhel9Affected
Red Hat Enterprise Linux AI (RHEL AI) 3rhelai3/bootc-gaudi-rhel9Affected
11 more in the JSON and on Red Hat's page.

Contains Red Hat Security Data © Red Hat, Inc., licensed under CC BY 4.0. Source: https://access.redhat.com/security/data. Modified: reformatted. Red Hat's page for CVE-2026-116177.

Attack patterns (CAPEC)

  • CAPEC-147 XML Ping of the Death Likelihood: Low · Severity: Medium
  • CAPEC-227 Sustained Client Engagement
  • CAPEC-492 Regular Expression Exponential Blowup

The attack patterns MITRE's CAPEC relates to this record's weaknesses, as MITRE names them. Credits.

References

Check your stackJSON