CVE-2026-115912: vLLM arbitrary code execution in shared-memory multimodal cache

CVE-2026-115912 · Severity: high · CVSS 7.5 · Published 2026-10-26

vLLM is an inference engine for large language models that can cache multimodal data using shared memory for performance. An authenticated attacker can forge cache references and cause the engine to deserialize malicious data with pickle, executing arbitrary Python code on the server. This affects scale-out deployments with shared-memory multimodal caching enabled.

Technologies: vllm (PyPI), vLLM Project vllm. Vendors: PyPI, vLLM Project.

Quick answers

Is CVE-2026-115912 exploited in the wild?
No. Junglewise Threat Intelligence has no report of CVE-2026-115912 being exploited in the wild.
How severe is CVE-2026-115912?
High severity, with a CVSS score of 7.5, published 26 October 2026.
Which products and versions are affected?
vllm (PyPI) < 0.32.0, ECOSYSTEM:introduced=0,fixed=0.32.0; vLLM Project vllm before 0.32.0.
Is there a fix for CVE-2026-115912?
Yes. vllm (PyPI): upgrade to 0.32.0 or later. Junglewise finds every place you run it, drafts the upgrade and applies it once you approve.
How is CVE-2026-115912 attacked?
It is attacked over the network, in a software library.
Am I affected by CVE-2026-115912?
Check whether you run vllm (PyPI) or vLLM Project vllm at an affected version. Junglewise does that check for you around the clock: connect your code, cloud and domains, and it matches new vulnerabilities like this one against what you actually run and tells you when one reaches you.

How to fix

  • vllm (PyPI): upgrade to 0.32.0 or later
    pip install "vllm>=0.32.0"

Technical details

An authenticated caller can place forged shared-memory addresses and monotonic_id values in multimodal feature kwargs_data. In scale-out deployments with shared-memory multimodal caching enabled, workers trust these values without validation and pass attacker-selected bytes to pickle.loads(), leading to arbitrary code execution even when insecure serialization is disabled. The fix adds HMAC-based authentication of cache handles using a server-side secret.

Affected products

  • vLLM Project vLLM before 0.32.0

Identifiers

  • CVE-2026-115912
  • GHSA-6757-hggm-hp2w
  • PYSEC-2026-4345

Timeline

  • 2026-10-26: disclosed
  • 2026-09-30: patched

References

This page as Markdown. Free to reuse under CC BY 4.0 with a link back.