CVE-2026-126398
Description
A flaw was found in vllm. A remote attacker could cause a Denial of Service (DoS) by triggering an internal assertion failure. When the service is configured to run an XD-RoPE model with prompt embeddings enabled, submitting an inference request containing only prompt embeddings without token identifiers causes the engine and API server to terminate unexpectedly. This crash disconnects all active clients and halts service availability until the server is restarted.
Statement
This vulnerability is rated as Moderate severity because exploitation depends on non-default operational prerequisites and impacts service availability without exposing sensitive data or allowing unauthorized modifications. In Red Hat environments, vLLM instances are vulnerable only when hosting architectures utilizing XD-RoPE scaling with prompt embeddings enabled under the V1 model runner. Submitting an inference payload containing prompt embeddings without accompanying token identifiers forces an internal assertion failure, resulting in the termination of the serving instance and disruption of ongoing requests.
Mitigation
Disable the prompt embeddings feature on affected vLLM instances by omitting or removing the `--enable-prompt-embeds` flag from the server startup configuration. Additionally, limit network exposure by restricting access to the vLLM serving port to trusted clients using network packet filters or an API reverse proxy.
For deployments managed via systemd:
1. Edit the service configuration to remove `--enable-prompt-embeds` from the execution command line:
systemctl edit vllm.service
2. Reload and restart the service:
systemctl daemon-reload
systemctl restart vllm.service
Caveats:
Disabling `--enable-prompt-embeds` will cause requests that provide prompt embeddings directly instead of standard text or token identifiers to be rejected.
Warning: Restarting or reloading the vLLM service terminates the running process and will interrupt any active client inference sessions, resulting in temporary service downtime.
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 Hat | NVD | cve.org | |
|---|---|---|---|
| Base Score | 5.9 | N/A | 5.9 |
| Attack Vector | Network | N/A | Network |
| Attack Complexity | High | N/A | High |
| Privileges Required | None | N/A | None |
| User Interaction | None | N/A | None |
| Scope | Unchanged | N/A | Unchanged |
| Confidentiality | None | N/A | None |
| Integrity Impact | None | N/A | None |
| Availability Impact | High | N/A | High |
Vector
Red Hat: CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:N/I:N/A:H
cve.org: CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:N/I:N/A:H
Understanding the Weakness (CWE)
Availability
Technical Impact: DoS: Crash, Exit, or Restart
While assertion is good for catching logic errors and reducing the chances of reaching more serious vulnerability conditions, an attacker can still cause a denial of service if they can trigger an assert statement, and if the scope of the assert() extends beyond the attacker's own session. For example, if a server handles multiple simultaneous connections, if an assert() occurs in one single connection that causes all other connections to be dropped, this could be a significant denial of service.
Frequently Asked Questions
Not sure what something means? Check out our Security Glossary.
Want to get errata notifications? Sign up here.