Service delay via unbudgeted logprobs computation
Published May 26, 2026 · Updated May 26, 2026
Denial of service in vLLM 0.19.0 allows remote attackers to delay co-scheduled requests through the OpenAI-compatible API. The V1 scheduler batches requests by token and sequence counts but does not budget sampling work, so high n and logprobs values multiply full-vocabulary Top-K sorting during every synchronous decode step. A remote client able to submit completion requests can impose 76-to-423-fold first-token latency on concurrent requests; the demonstrated effect is severe delay rather than a crash.
Summary
What happened
Denial of service in vLLM 0.19.0 allows remote attackers to delay co-scheduled requests through the OpenAI-compatible API. The V1 scheduler batches requests by token and sequence counts but does not budget sampling work, so high n and logprobs values multiply full-vocabulary Top-K sorting during every synchronous decode step. A remote client able to submit completion requests can impose 76-to-423-fold first-token latency on concurrent requests; the demonstrated effect is severe delay rather than a crash.
The record
- CVE
- CVE-2026-9540
- Published
- May 26, 2026
- Updated
- May 26, 2026
- Vendor
- vLLM Project
- Product
- vLLM
- Classifications
- CWE-404, T1499
- Attack vector
- network
- Privileges
- unauthenticated
Timeline
How it unfolded
- May 26, 2026CVE publishedPublication date reported by the CVE source.
- May 26, 2026Record updatedLatest update available in the CVE record.
Exploitability
Present is not the same as exploitable
Compare your product and version with the public record. A matching version still requires validation against your environment.
Is a vulnerable build present?
Compare these published version ranges with your installed build and any vendor patches.
- Affected versionversion=0.19.0
What conditions does exploitation require?
What is affected?
Attacks
What attackers are doing with it
Daily unique IPs observed by Shadowserver honeypots for known exploited vulnerabilities (KEVs). Missing observations do not establish an absence of attacks.
Weakness, pattern, technique
Public exploit references
- vLLM issue 37343 reproduction scriptproof of concept · demonstrated
Labels summarize the accepted research assessment. They do not indicate a test against your environment.
Technologies
Your stack
See the directory against your own environment.
Your stack
Check the software in your environment
Book a demo to see how Hinoki identifies affected software and validates exploitability in your environment.
Book a demo