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.

CVSS 3.1

7.5CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H

EPSS

0.6% chance of exploitation in the next 30 days; higher than 49% of CVEs (scores of 2026-10-25)

CWEs

CWE-20, CWE-1284

Published

2026-07-06

Modified

2026-10-25

Record from

the NVD, synced 2026-10-25

Affected

• vllm:vllm

Packages (OSV)

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
• https://github.com/vllm-project/vllm/security/advisories/GHSA-8wr5-jm2h-8r4f

Check your stackJSON