vLLM through 0.29.0 fails to properly clean up decode-side metadata for rejected inference requests in prefill/decode disaggregated deployments. Remote attackers can submit requests with max_tokens=0 to exhaust decode-worker memory without bound until the worker restarts.
References
Configurations
History
28 Sep 2026, 18:37
| Type | Values Removed | Values Added |
|---|---|---|
| First Time |
Vllm vllm
Vllm |
|
| CPE | cpe:2.3:a:vllm:vllm:*:*:*:*:*:*:*:* | |
| References | () https://github.com/vllm-project/vllm - Product | |
| References | () https://github.com/vllm-project/vllm/blob/v0.29.0/vllm/distributed/kv_transfer/kv_connector/v1/nixl/push_worker.py#L162-L181 - Product | |
| References | () https://github.com/vllm-project/vllm/pull/55677 - Issue Tracking, Patch | |
| References | () https://www.vulncheck.com/advisories/vllm-through-0.29.0-memory-exhaustion-via-rejected-requests - Third Party Advisory |
17 Sep 2026, 23:18
| Type | Values Removed | Values Added |
|---|---|---|
| New CVE |
Information
Published : 2026-09-17 23:18
Updated : 2026-09-28 18:37
NVD link : CVE-2026-93436
Mitre link : CVE-2026-93436
CVE.ORG link : CVE-2026-93436
JSON object : View
Products Affected
vllm
- vllm
CWE
CWE-401
Missing Release of Memory after Effective Lifetime
