Open-Source Security Intelligence

Know every vulnerability
before it knows you.

DevGuard continuously monitors your dependencies and alerts you when CVEs like this one affect your stack — with real-time threat intelligence built for developers.

Search

GHSA-h3rc-6mm3-gc2m

HighCVSS 8.1 / 10
Published Oct 27, 2026·Last modified Oct 27, 2026
Affected Components(1)
PyPI logovllm
0.7.3 – 0.31.0
Description

Summary

An untrusted multimodal request can set mm_processor_kwargs.code_revision. When vLLM serves a model whose Hugging Face processor uses dynamic auto_map code and the operator enabled --trust-remote-code, this request value reaches AutoProcessor.from_pretrained and selects the Python revision imported by Transformers.

This lets a client replace the operator-reviewed processor module with code from another revision of the same model repository. The operator-selected model/data revision remains pinned; only the executed processor code changes.

Importing a Python module is arbitrary code execution in the vLLM API/renderer process. The PoC below demonstrates this with an import-time filesystem write. It is fully offline and does not require a GPU, model weights, or changes to a real Hub repository.

Affected snapshot

Affected configuration

The vulnerable path exists when all of the following are true:

  1. The served model's processor uses dynamic auto_map.AutoProcessor code.
  2. The operator enabled trust_remote_code.
  3. The server can fetch the request-selected revision, or it is already cached.
  4. The attacker can create or otherwise control a revision in the served model repository. An example is a Hugging Face community pull-request ref when that repository permits them.
  5. A multimodal request reaches the processor path.

The default trust_remote_code=False configuration is not affected.

Attack surface

mm_processor_kwargs is accepted directly by public request models:

/tokenize is particularly relevant because vLLM's authentication middleware guards only /v1, /v2, /inference, and /cohere prefixes:

Consequently, /tokenize remains unauthenticated even when --api-key is configured, unless an external proxy protects it. The vLLM Security Guide documents /tokenize as an unprotected endpoint.

Root cause

The request value follows this path:

  1. OnlineRenderer.preprocess_chat places request mm_processor_kwargs in prompt extras.
  2. Multimodal rendering forwards those kwargs into processor construction through MMProcessorInputs and _process_multimodal_async.
  3. MultiModalConfig.merge_mm_processor_kwargs overlays inference values on startup values, so the request wins.
  4. cached_get_processor_without_dynamic_kwargs filters only recognized processor-call keys. code_revision is retained as a processor-construction kwarg.
  5. cached_processor_from_config pins revision and trust_remote_code, but forwards the remaining kwargs without reserving code_revision.
  6. get_processor calls AutoProcessor.from_pretrained(..., revision=revision, trust_remote_code=trust_remote_code, **kwargs).
  7. Transformers AutoProcessor.from_pretrained uses code_revision to retrieve and import the dynamic module through get_class_from_dynamic_module, then removes it before loading processor data. This is why code comes from the attacker revision while configuration/data remain operator-pinned.

Request-time processor arguments are not separated from processor-construction arguments, and code_revision is not reserved.

Suggested remediation

  1. Separate processor-construction arguments from untrusted processor-call arguments. An allowlist of supported request-time preprocessing controls is safest.
  2. Reject code_revision and other loader/source controls in request mm_processor_kwargs, and pass model_config.code_revision explicitly as the only permitted processor-code revision.
  3. Add a regression test proving a request cannot change the loaded module revision while operator pins remain effective.
  4. Protect /tokenize with the configured authentication policy as defense in depth.

Expected safe behavior is either a 4xx response for reserved request keys or silently ignoring them before processor construction. Rejection is preferable because it makes attempted misuse visible.

A patch for this was merged here: https://github.com/vllm-project/vllm/pull/58830

Upload your SBOM

Upload your own SBOM in CycloneDX 1.6 or higher (JSON) directly here to check your vulnerabilities.

Risk Scores
Base Score
8.1

The vulnerability can be exploited over the network without needing physical access. It is difficult for an attacker to exploit this vulnerability and may require special conditions. An attacker does not need any special privileges or access rights. No user interaction is needed for the attacker to exploit this vulnerability. The impact is confined to the system where the vulnerability exists. There is a high impact on the confidentiality of the information. There is a high impact on the integrity of the data. There is a high impact on the availability of the system.

Threat Intelligence
7.4

Exploitation activity has been observed. Apply available patches or mitigations urgently.

EPSS
0.46%

The exploit probability is very low. The vulnerability is unlikely to be exploited in the next 30 days.

Exploit
Not available

We did not find any exploit available. Neither in GitHub repositories nor in the Exploit-Database.

Browse More

Scan your project

Continuously monitor your dependencies and get alerted when vulnerabilities like this one affect your stack.

Checkout DevGuard