Allocation of Resources Without Limits or Throttling Affecting vllm package, versions [,0.31.0)
Threat Intelligence
The probability is the direct output of the EPSS model, and conveys an overall sense of the threat of exploitation in the wild. The percentile measures the EPSS probability relative to all known EPSS scores. Note: This data is updated daily, relying on the latest available EPSS model version. Check out the EPSS documentation for more details.
Do your applications use this vulnerable package?
In a few clicks we can analyze your entire application and see what components are vulnerable in your application, and suggest you quick fixes.
Test your applicationsSnyk Learn
Learn about Allocation of Resources Without Limits or Throttling vulnerabilities in an interactive lesson.
Start learning- Snyk IDSNYK-PYTHON-VLLM-21347784
- published21 Oct 2026
- disclosed21 Oct 2026
- creditDOHA LEE
Introduced: 21 Oct 2026
NewCVE-2026-114052 (opens in a new tab)How to fix?
Upgrade vllm to version 0.31.0 or higher.
Overview
vllm is an A high-throughput and memory-efficient inference and serving engine for LLMs
Affected versions of this package are vulnerable to Allocation of Resources Without Limits or Throttling via MossAudioProcessingInfo.get_hf_processor() in MOSS-Audio model serving, which caches a MossAudioProcessor instance for every distinct request-controlled mm_processor_kwargs value in an unbounded private dictionary (_hf_processor_cache) with no capacity limit, TTL, or eviction policy. An unauthenticated attacker can send requests to the /tokenize endpoint - which is intentionally excluded from API-key-protected route prefixes, bypassing --api-key - with distinct or oversized mel_config values (e.g. mel_dim and mel_n_fft) to trigger persistent linear memory growth or a single catastrophically large allocation, exhausting host memory and crashing the API process. Each cached processor contains a WhisperFeatureExtractor with an allocated mel-filter array, and the approximate allocation is (floor(mel_n_fft / 2) + 1) * mel_dim * 8 bytes with no enforced upper bound on either dimension.
Note: This is only exploitable when the vLLM instance is serving a MOSS-Audio model.