CVE Catalog

CVE-2026-94627

HighCVSS 7.5
Published: Updated: Translated: NVD NIST

Exploitation Probability (EPSS)

Low risk
0.63%

48th percentile - higher than 48% of all known CVEs

Summary

vLLM Mooncake connector through 0.29.0 fails to properly manage GPU KV cache block ownership when concurrent child requests share a single transfer ID in prefill/decode disaggregated deployments. Attackers can trigger GPU memory exhaustion by submitting completion requests with multiple prompts, causing orphaned KV cache blocks to accumulate until process restart and eventually preventing legitimate requests from executing.

Risk Assessment

An attacker can cause GPU memory exhaustion and denial of service for legitimate requests, forcing a process restart. This affects vLLM deployments using the Mooncake connector in prefill/decode disaggregated mode.

Recommendation

Upgrade vLLM and the Mooncake connector to a version containing the fix, or restrict access to completion endpoints for untrusted users. Monitor GPU memory usage and unusual request patterns.

Related vulnerabilities

Original NVD description (English source)

vLLM Mooncake connector through 0.29.0 fails to properly manage GPU KV cache block ownership when concurrent child requests share a single transfer ID in prefill/decode disaggregated deployments. Attackers can trigger GPU memory exhaustion by submitting completion requests with multiple prompts, causing orphaned KV cache blocks to accumulate until process restart and eventually preventing legitimate requests from executing.

Vulnerability data from NVD (NIST) · CISA KEV · EPSS