High severity7.5NVD Advisory· Published Sep 17, 2026· Updated Sep 28, 2026

CVE-2026-93436

Description

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.

AI Insight

LLM-synthesized narrative grounded in this CVE's description and references.

Affected products

Vllm/Vllmllm-create2 versions
<=0.29.0+ 1 more
- (no CPE)range: <=0.29.0
- cpe:2.3:a:vllm:vllm:*:*:*:*:*:*:*:*range: <=0.29.0

Patches

Vulnerability mechanics

References

github.com/vllm-project/vllm/pull/55677nvdIssue TrackingPatch
www.vulncheck.com/advisories/vllm-through-0.29.0-memory-exhaustion-via-rejected-requestsnvdThird Party Advisory
github.com/vllm-project/vllm/blob/v0.29.0/vllm/distributed/kv_transfer/kv_connector/v1/nixl/push_worker.pynvdProduct

News mentions

VLLM: Batch of 17 Vulnerabilities Disclosed, Affecting LLM Inference Engine Stability
Vypr Intelligence · Sep 21, 2026

cvss	0.488
epss	0.001
exploit	0.000
kev	0.000
patch	-0.070
ransomware	0.000