CVE-2026-114052: vLLM 0.25.0 before 0.31.0 MOSS-Audio Memory Exhaustion via /tokenize mm_processor_kwargs
vLLM after 0.24.0 and before 0.31.0 contains a memory exhaustion vulnerability in MOSS-Audio model serving that caches a processor for every distinct request-controlled mmprocessorkwargs value in an unbounded dictionary. Unauthenticated attackers can send distinct or oversized melconfig values to the /tokenize endpoint, which bypasses --api-key, to exhaust host memory and crash the API process.
Affected Software
Remediation
Recommended actions to resolve this vulnerability, in priority order.
- Upgrade
Upgrade
pip/vllmto a version that resolves this vulnerability.Fixed in 0.31.0 - Upgrade
Upgrade
vLLMto a version that resolves this vulnerability.Fixed in 0.31.0
Event History
Frequently Asked Questions
Which deployments are exposed?
Deployments serving the MOSS-Audio model with vLLM versions after 0.24.0 and before 0.31.0 are affected. The exposed endpoint is /tokenize.
Does an attacker need credentials or an API key?
No. The /tokenize endpoint bypasses --api-key, so an unauthenticated network attacker can send requests to trigger the issue.
What request behavior triggers the denial of service?
An attacker can submit distinct or oversized mel_config values through mm_processor_kwargs. Each distinct value causes a processor to be cached in an unbounded dictionary, consuming host memory until the API process crashes.
What should be prioritized for mitigation?
Upgrade to vLLM 0.31.0 or later. Until upgrading, restrict untrusted access to /tokenize and prevent unauthenticated clients from reaching the API service.