vLLM: Request-controlled mm_processor_kwargs.code_revision allows remote code execution

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

- Repository: [vllm-project/vllm](https://github.com/vllm-project/vllm) - Verified commit: [570735520942b6428ae98f1eb442c6d88adae55e](https://github.com/vllm-project/vllm/commit/570735520942b6428ae98f1eb442c6d88adae55e) - Verified with Transformers 5.15.0 and 5.16.1 - Affected release range has not been bisected

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](https://huggingface.co/docs/hub/en/repositories-pull-requests-discussions#where-in-the-git-repo-are-changes-stored) 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:

- [TokenizeChatRequest.mm_processor_kwargs](https://github.com/vllm-project/vllm/blob/570735520942b6428ae98f1eb442c6d88adae55e/vllm/entrypoints/serve/tokenize/protocol.py#L111-L114) - [ChatCompletionRequest.mm_processor_kwargs](https://github.com/vllm-project/vllm/blob/570735520942b6428ae98f1eb442c6d88adae55e/vllm/entrypoints/openai/chat_completion/protocol.py#L372-L375)

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

- [AuthenticationMiddleware](https://github.com/vllm-project/vllm/blob/570735520942b6428ae98f1eb442c6d88adae55e/vllm/entrypoints/serve/middleware/authenticate.py#L11-L62)

Consequently, /tokenize remains unauthenticated even when --api-key is configured, unless an external proxy protects it. The vLLM [Security Guide](https://docs.vllm.ai/en/latest/usage/security/#unprotected-endpoints-no-api-key-required) documents /tokenize as an unprotected endpoint.

Root cause

The request value follows this path:

1. [OnlineRenderer.preprocess_chat](https://github.com/vllm-project/vllm/blob/570735520942b6428ae98f1eb442c6d88adae55e/vllm/renderers/online_renderer.py#L406-L435) places request mm_processor_kwargs in prompt extras. 2. Multimodal rendering forwards those kwargs into processor construction through [MMProcessorInputs](https://github.com/vllm-project/vllm/blob/570735520942b6428ae98f1eb442c6d88adae55e/vllm/renderers/base.py#L751-L760) and [_process_multimodal_async](https://github.com/vllm-project/vllm/blob/570735520942b6428ae98f1eb442c6d88adae55e/vllm/renderers/base.py#L831-L847). 3. [MultiModalConfig.merge_mm_processor_kwargs](https://github.com/vllm-project/vllm/blob/570735520942b6428ae98f1eb442c6d88adae55e/vllm/config/multimodal.py#L506-L518) overlays inference values on startup values, so the request wins. 4. [cached_get_processor_without_dynamic_kwargs](https://github.com/vllm-project/vllm/blob/570735520942b6428ae98f1eb442c6d88adae55e/vllm/transformers_utils/processor.py#L344-L375) filters only recognized processor-call keys. code_revision is retained as a processor-construction kwarg. 5. [cached_processor_from_config](https://github.com/vllm-project/vllm/blob/570735520942b6428ae98f1eb442c6d88adae55e/vllm/transformers_utils/processor.py#L380-L390) pins revision and trust_remote_code, but forwards the remaining kwargs without reserving code_revision. 6. [get_processor](https://github.com/vllm-project/vllm/blob/570735520942b6428ae98f1eb442c6d88adae55e/vllm/transformers_utils/processor.py#L205-L245) calls AutoProcessor.from_pretrained(..., revision=revision, trust_remote_code=trust_remote_code, **kwargs). 7. Transformers [AutoProcessor.from_pretrained](https://github.com/huggingface/transformers/blob/v5.15.0/src/transformers/models/auto/processing_auto.py#L294-L302) uses code_revision to retrieve and import the dynamic module through [get_class_from_dynamic_module](https://github.com/huggingface/transformers/blob/v5.15.0/src/transformers/dynamic_module_utils.py#L474-L575), 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 [...]