CVE-2026-54234

The record

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.

Affected packages

Related vulnerabilities

• CVE-2026-113786 pypdf is a free and open-source pure-python PDF library. CWE-20
• CVE-2026-112081 A vulnerability was determined in hankcs HanLP up to 2.1.0-beta.64. CWE-20
• CVE-2026-111869 A vulnerability was identified in makeplane plane up to 1.4.2. CWE-20
• CVE-2026-100347 kin-openapi is a Go project for handling OpenAPI files. CWE-20
• CVE-2026-99999 In the Linux kernel, the following vulnerability has been resolved: ipv4: validate… CWE-20
• CVE-2026-107726 Hazelcast is a unified real-time data platform combining stream processing with a fast… CWE-20
• CVE-2026-105757 vLLM is an inference and serving engine for large language models. CWE-20
• CVE-2026-105756 vLLM is an inference and serving engine for large language models. CWE-20