CVE-2026-126395

Description

A flaw was found in vLLM. The error sanitization routine fails to redact uppercase memory addresses from error messages during image processing. An unauthenticated remote attacker can exploit this vulnerability by submitting a crafted request that triggers an error response disclosing internal heap memory addresses. This information disclosure can allow an attacker to bypass Address Space Layout Randomization (ASLR) security defenses.

Statement

This vulnerability is evaluated as having a Moderate severity because it leads exclusively to partial information disclosure without enabling direct remote code execution or state alteration. While remote callers can prompt verbose error messages revealing runtime heap addresses on multimodal endpoints, the flaw does not leak raw model weights, confidential user inputs, or process memory contents. Such heap address disclosures typically present an operational risk only when leveraged to overcome address space layout randomization (ASLR) in conjunction with an independent memory corruption defect. In standard Red Hat AI environments, these endpoints are frequently restricted to authenticated workloads or isolated within private networks, substantially mitigating external exposure.

Mitigation

Restrict network exposure and filter error responses returned by the vLLM service:

1. Restrict API access to trusted networks:
Use firewalld to limit network access to the vLLM service port (default TCP port 8000) to ensure only authorized clients or reverse proxies can connect:

firewall-cmd --permanent --zone=public --remove-port=8000/tcp
firewall-cmd --permanent --zone=trusted --add-source=<trusted_ip_or_subnet>
firewall-cmd --reload

2. Intercept and mask error payloads:
Position a reverse proxy or API gateway ahead of the vLLM instance configured to substitute generic response bodies for HTTP 4xx and 5xx status codes instead of returning upstream application error strings.

Caveats: Restricting network reachability prevents unauthorized users from accessing the inference service. Suppressing verbose error messages may complicate client-side debugging when malformed requests are processed.

Warning: Modifying firewall configurations and reloading the firewall service can disrupt active network sessions.

Common Vulnerability Scoring System (CVSS) Score Details

Info alert:Important note

CVSS scores for open source components depend on vendor-specific factors (e.g. version or build chain). Therefore, Red Hat's score and impact rating can be different from NVD and other vendors. Red Hat remains the authoritative CVE Naming Authority (CNA) source for its products and services (see Red Hat classifications).

The following CVSS metrics and score provided are preliminary and subject to review.

CVSS v3 Score Breakdown

Red HatNVDcve.org
Base Score5.3N/A5.3
Attack VectorNetworkN/ANetwork
Attack ComplexityLowN/ALow
Privileges RequiredNoneN/ANone
User InteractionNoneN/ANone
ScopeUnchangedN/AUnchanged
ConfidentialityLowN/ALow
Integrity ImpactNoneN/ANone
Availability ImpactNoneN/ANone

Vector

Red Hat: CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:N/A:N

cve.org: CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:N/A:N

Understanding the Weakness (CWE)

Confidentiality

Technical Impact: Read Application Data

Often this will either reveal sensitive information which may be used to launch another, more focused attack or disclose private information stored in the server. For example, an attempt to exploit a path traversal weakness (CWE-22) might yield the full pathname of the installed application. In turn, this could be used to select the proper number of ".." sequences to navigate to the targeted file. An attack using SQL injection (CWE-89) might not initially succeed, but an error message could reveal the malformed query, which would expose query logic and possibly even passwords or other sensitive information used within the query.

Frequently Asked Questions

Want to get errata notifications? Sign up here.