CVE-2026-93436: vLLM through 0.29.0 Memory Exhaustion via Rejected Requests
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 maxtokens=0 to exhaust decode-worker memory without bound until the worker restarts.
Affected Software
Event History
Frequently Asked Questions
Which deployments are exposed to this issue?
The issue affects vLLM through 0.29.0 when it is deployed with prefill/decode disaggregation. The unbounded memory growth occurs on decode workers.
What does an attacker need to do to trigger the exhaustion?
A remote attacker can submit inference requests with max_tokens=0. No privileges or user interaction are required according to the supplied severity vector.
What is the operational impact?
Rejected requests can leave decode-side metadata allocated, causing decode-worker memory consumption to grow without bound. The worker may become unavailable until it restarts.