Authenticated service crash via proposer shape mismatch
Published May 12, 2026 · Updated May 12, 2026
Denial of service in vLLM 0.18.0 through 0.19.1 allows remote authenticated users to crash EngineCore via a request with sampling penalties. The extract_hidden_states proposer returns a two-column sampled-token tensor after the first decode step, but penalty processing expects one column and raises a broadcast-shape RuntimeError. The server must use extract_hidden_states speculative decoding; one request with a repetition, frequency, or presence penalty then causes complete service interruption.
Summary
What happened
Denial of service in vLLM 0.18.0 through 0.19.1 allows remote authenticated users to crash EngineCore via a request with sampling penalties. The extract_hidden_states proposer returns a two-column sampled-token tensor after the first decode step, but penalty processing expects one column and raises a broadcast-shape RuntimeError. The server must use extract_hidden_states speculative decoding; one request with a repetition, frequency, or presence penalty then causes complete service interruption.
The record
- CVE
- CVE-2026-44223
- Published
- May 12, 2026
- Updated
- May 12, 2026
- Vendor
- vLLM Project
- Product
- vLLM
- Classifications
- CWE-131, CWE-704, T1499.004
- Attack vector
- network
- Privileges
- authenticated
Timeline
How it unfolded
- May 12, 2026CVE publishedPublication date reported by the CVE source.
- May 12, 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=>= 0.18.0, < 0.20.0
What conditions does exploitation require?
What is affected?
Published CVSS scores
CVSS describes severity. EPSS estimates exploitation probability.
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
- Single-request sampling-penalty crash triggerfunctional · 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