CVE-2026-54234

Description

vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. Prior to 0.24.0, a frontend-legal multi-request speculative decoding workload can cause the rejection sampler to produce a recovered token equal to the model vocabulary size boundary value, which is then converted to negative one when the engine selects the next live token for a request and is written back into the drafter's input ids; that out-of-vocabulary value is later consumed by the model's embedding and attention path and crashes the engine worker with a GPU device-side assertion. The same triggering request sequence is reachable through the public gRPC Generate and Abort endpoints, so a remote client that can send generation requests can crash the shared engine worker, aborting concurrent requests and causing a service-wide denial of service for other clients of the deployment until the worker is restarted. This issue is fixed in version 0.24.0.

CVSS v3.1 Score

7.5

HIGH

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

Weakness Type (CWE)

CWE-20 Improper Input Validation

CWE-1284 CWE-1284

References

Other References

https://github.com/vllm-project/vllm/commit/8a5cf1ccd65e8ac7635c402c1ec0b08988bc26ca https://github.com/vllm-project/vllm/pull/44744 https://github.com/vllm-project/vllm/security/advisories/GHSA-8wr5-jm2h-8r4f

Frequently Asked Questions

What is CVE-2026-54234? +

vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. Prior to 0.24.0, a frontend-legal multi-request speculative decoding workload can cause the rejection sampler to produce a recovered token equal to the model vocabulary size boundary value, which is then converted to negative one when the engine selects the next live token for a request and is written back into the drafter's input ids; that out-of-vocabulary value is later consumed by the model's embedding and attention path and crashes the engine worker with a GPU device-side assertion. The same triggering request sequence is reachable through the public gRPC Generate and Abort endpoints, so a remote client that can send generation requests can crash the shared engine worker, aborting concurrent requests and causing a service-wide denial of service for other clients of the deployment until the worker is restarted. This issue is fixed in version 0.24.0. It has a CVSS v3.1 base score of 7.5 (HIGH).

How severe is CVE-2026-54234? +

CVE-2026-54234 has a CVSS v3.1 score of 7.5 out of 10, rated HIGH. This is a high-severity vulnerability that should be prioritized for patching.

How do I check if I'm vulnerable to CVE-2026-54234? +

You can use Secably's free Website Scanner to check your website for known vulnerabilities. For infrastructure scanning, use the Port Scanner to identify exposed services that may be affected. Check the vendor advisories linked above for specific patch and version information.

Related Vulnerabilities

CVE-2025-13826

Zervit's portable HTTP/web server is vulnerable to remote DoS attacks when a configuration reset request is made. The vulnerability is …

CVE-2024-54011

Penetration Testing engineers at Amazon have discovered a flaw where the camera system fails to properly handle data supplied in …

CVE-2026-6409

A Denial of Service (DoS) vulnerability exists in the Protobuf PHP library during the parsing of untrusted input. Maliciously structured …

CVE-2026-12569

A critical remote code execution (RCE) vulnerability has been reported in PTC Windchill PDMlink and PTC FlexPLM. The vulnerability may …

CVE-2025-23202

Bible Module is a tool designed for ROBLOX developers to integrate Bible functionality into their games. The `FetchVerse` and `FetchPassage` …

CVE-2025-24501

An improper input validation allows an unauthenticated attacker to alter PAM logs by sending a specially crafted HTTP request.