CVE-2026-54234
CVE-2026-54234 is a high-severity vulnerability in Vllm with a CVSS 3.x base score of 7.5. It is not currently listed as actively exploited by CISA, and its EPSS exploit-prediction score is low. The underlying weakness is classified as CWE-20.
Key facts
- Severity: High (CVSS 3.x base score 7.5)
- EPSS exploit prediction: 0% (26th percentile)
- Actively exploited: Not listed in CISA KEV
- Weakness: CWE-20
- Affected product: Vllm
- Published:
- Last modified:
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.
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.
- How severe is CVE-2026-54234?
- CVE-2026-54234 has a CVSS 3.x base score of 7.5, rated high severity. It is exploitable over network with low attack complexity, requires no privileges and no user interaction. Impact on confidentiality is none, integrity none, and availability high.
- Is CVE-2026-54234 being actively exploited?
- It is not currently listed in CISA's KEV catalog. Its EPSS exploit-prediction score is 0% (26th percentile), an estimate of the probability of exploitation in the next 30 days.
- What products are affected by CVE-2026-54234?
- CVE-2026-54234 affects Vllm. See the affected-products list for the exact vulnerable versions.
- How do I fix CVE-2026-54234?
- Review the linked vendor and NVD advisories for patched versions and mitigations, then upgrade or apply the recommended workaround. Given its high severity, prioritise patching exposed systems.
- When was CVE-2026-54234 published?
- CVE-2026-54234 was published on 2026-07-06 and last updated on 2026-07-07.
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
Affected products (1)
- cpe:2.3:a:vllm:vllm:*:*:*:*:*:*:*:*
More vulnerabilities in Vllm
- CVE-2025-32444 — Critical (CVSS 10.0): vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. Versions starting from 0.6.5 and…
- CVE-2026-22778 — Critical (CVSS 9.8): vLLM is an inference and serving engine for large language models (LLMs). From 0.8.3 to before 0.14.1, when an invalid…
- CVE-2025-47277 — Critical (CVSS 9.8): vLLM, an inference and serving engine for large language models (LLMs), has an issue in versions 0.6.5 through 0.8.4…
- CVE-2024-11041 — Critical (CVSS 9.8): vllm-project vllm version v0.6.2 contains a vulnerability in the MessageQueue.dequeue() API function. The function uses…
- CVE-2026-48746 — Critical (CVSS 9.1): vLLM is an inference and serving engine for large language models (LLMs). From 0.3.0 until 0.22.0, a vulnerability in…
- CVE-2025-29783 — Critical (CVSS 9.0): vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. When vLLM is configured to use…
Other CWE-20 (Improper Input Validation) vulnerabilities
- CVE-2026-48316 — Critical (CVSS 10.0): ColdFusion versions 2025.9, 2023.20 and earlier are affected by an Improper Input Validation vulnerability that could…
- CVE-2026-48281 — Critical (CVSS 10.0): ColdFusion versions 2025.9, 2023.20 and earlier are affected by an Improper Input Validation vulnerability that could…
- CVE-2026-48277 — Critical (CVSS 10.0): ColdFusion versions 2025.9, 2023.20 and earlier are affected by an Improper Input Validation vulnerability that could…
- CVE-2026-48055 — Critical (CVSS 10.0): Streambert is a cross-platform Electron Desktop App to stream and download any video media. In versions 2.4.0 and…
- CVE-2026-34910 — Critical (CVSS 10.0): A malicious actor with access to the network could exploit an Improper Input Validation vulnerability found in UniFi OS…
- CVE-2026-33587 — Critical (CVSS 10.0): Lack of user input sanitisation in Open Notebook v1.8.3 allows the application user to execute Python code (and…
Browse all CWE-20 (Improper Input Validation) vulnerabilities →