CVE-2026-73559
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]], prompt_to_seq() in vllm/renderers/inputs/preprocess.py and OnlineRenderer.preprocess_completion() in vllm/renderers/online_renderer.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.
Properties
- summary
- vLLM: Completion prompt lists fan out into unbounded engine requests
- severity
- MEDIUM
- cvss_severity
- MEDIUM
- epss_score
- 0.00548
- cvss_score
- 6.5
- retrieved_at
- 2026-09-27T13:53:59+00:00
- ghsa_published
- 2026-08-13T18:40:06Z
- source_url
- https://github.com/advisories/GHSA-87x5-vmc3-756j
- ghsa_updated
- 2026-08-13T18:40:09Z
- ghsa_id
- GHSA-87x5-vmc3-756j
- last_source
- FIRST EPSS
- score
- 6.5
- cve_id
- CVE-2026-73559
- cvss_vector
- CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H
- signal_observed_at
- 2026-09-11T17:54:58+00:00
- is_ghsa_only
- false
- vector
- CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H
- published_at
- 2026-08-13T16:19:05.863
- last_modified
- 2026-09-09T20:58:37.713
- epss_percentile
- 0.43662
Related Entities (6)
DESCRIBED_BY (1)
ENRICHED_BY (1)
VULNERABLE_TO (1)
AFFECTS (1)
HAS_WEAKNESS (1)
REPORTED_BY (1)
Explore deeper with Ninja Signal's threat intelligence graph