CVE-2026-73559: vLLM: Completion prompt lists fan out into unbounded engine requests
vLLM is an inference and serving engine for large language models. From 0.19.0 until 0.26.0, the /v1/completions CompletionRequest.prompt field in vllm/entrypoints/openai/completion/protocol.py accepts an unbounded list[str] or list[list[int]], prompttoseq() in vllm/renderers/inputs/preprocess.py and OnlineRenderer.preprocesscompletion() in vllm/renderers/onlinerenderer.py expand every element, and vllm/entrypoints/openai/completion/serving.py creates one engine generator and response slot per prompt, allowing an authenticated API client to exhaust CPU, memory, async scheduling capacity, engine request slots, and response buffering with one request. This issue is fixed in version 0.26.0.
Affected Software
Remediation
Recommended actions to resolve this vulnerability, in priority order.
- Upgrade
Upgrade
vllmto a version that resolves this vulnerability.Fixed in 0.26.0
Event History
Frequently Asked Questions
What is the severity of CVE-2026-73559?
CVE-2026-73559 has a medium severity rating of 6.5.
How do I fix CVE-2026-73559?
To fix CVE-2026-73559, update vLLM to version 0.26.1 or later.
What does CVE-2026-73559 affect?
CVE-2026-73559 affects the /v1/completions CompletionRequest.prompt field in versions 0.19.0 through 0.26.0 of vLLM.
What are the potential risks associated with CVE-2026-73559?
The risk associated with CVE-2026-73559 includes unbounded engine requests that could lead to resource exhaustion.
Is there a workaround for CVE-2026-73559?
Currently, the recommended action is to upgrade to a patched version, as there are no established workarounds for CVE-2026-73559.