CVE-2026-94627CWE-401

CVE-2026-94627

High · published September 22, 2026

CVSS v3.1
7.5
EPSS
1%
Percentile
49.5
In the wild
Unconfirmed
What it is

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.

The record
Technical detail
CVSS v3.1
7.5 · HIGH
Vector
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H
CVSS v4.0
8.7 · CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:N/VI:N/VA:H/SC:N/SI:N/SA:N
EPSS
0.00627 · 49.5th percentile
Weakness
CWE-401 · Missing Release of Memory after Effective Lifetime
Published
2026-09-22T02:17Z
Affected products (1)
ProductVersionsFixed in
vllm/vllm≤ 0.29.0—
References (4)
https://github.com/vllm-project/vllm · [email protected]https://github.com/vllm-project/vllm/blob/v0.29.0/vllm/distributed/kv_transfer/kv_connector/v1/mooncake/mooncake_connector.py#L1978-L1989 · [email protected]https://github.com/vllm-project/vllm/pull/49796 · [email protected]https://www.vulncheck.com/advisories/vllm-through-0.29.0-gpu-kv-cache-leak-via-mooncake-transfer-id-collision · [email protected]
EPSS history
Timeline
  • 23 SEP 03:49Z
    EPSS moved — → 0%
    epss
  • 21 SEP 22:04Z
    vLLM through 0.29.0 GPU KV Cache Leak via Mooncake Transfer ID Collision
    cvelistv5