VYPR
High severityNVD Advisory· Published Mar 26, 2026· Updated Mar 27, 2026

vLLM's hardcoded trust_remote_code=True in NemotronVL and KimiK25 bypasses user security opt-out

CVE-2026-27893

Description

vLLM is an inference and serving engine for large language models (LLMs). Starting in version 0.10.1 and prior to version 0.18.0, two model implementation files hardcode trust_remote_code=True when loading sub-components, bypassing the user's explicit --trust-remote-code=False security opt-out. This enables remote code execution via malicious model repositories even when the user has explicitly disabled remote code trust. Version 0.18.0 patches the issue.

Affected packages

Versions sourced from the GitHub Security Advisory.

PackageAffected versionsPatched versions
vllmPyPI
>= 0.10.1, < 0.18.00.18.0

Affected products

1

Patches

1
00bd08edeee5

[Security] Respect user trust_remote_code setting in NemotronVL and KimiK25 (#36192)

https://github.com/vllm-project/vllmRussell BryantMar 6, 2026via ghsa
2 files changed · +7 2
  • vllm/model_executor/models/kimi_k25.py+2 1 modified
    @@ -174,7 +174,8 @@ def __init__(self, ctx: InputProcessingContext) -> None:
             self.hf_config = self.get_hf_config()
             self.media_token_id = self.hf_config.media_placeholder_token_id
             media_processor = cached_get_image_processor(
    -            self.ctx.model_config.model, trust_remote_code=True
    +            self.ctx.model_config.model,
    +            trust_remote_code=self.ctx.model_config.trust_remote_code,
             )
             self.media_processor = media_processor
             self.hf_processor = MoonshotKimiVAutoProcessor(
    
  • vllm/model_executor/models/nemotron_vl.py+5 1 modified
    @@ -402,6 +402,7 @@ def __init__(self, *, vllm_config: VllmConfig, prefix: str = "") -> None:
             multimodal_config = vllm_config.model_config.multimodal_config
     
             self.config = config
    +        self.model_config = vllm_config.model_config
             self.multimodal_config = multimodal_config
             self._patch_quant_config(config, quant_config)
     
    @@ -456,7 +457,10 @@ def _init_vision_model(
             *,
             prefix: str,
         ):
    -        return AutoModel.from_config(config.vision_config, trust_remote_code=True)
    +        return AutoModel.from_config(
    +            config.vision_config,
    +            trust_remote_code=self.model_config.trust_remote_code,
    +        )
     
         def _init_mlp1(
             self,
    

Vulnerability mechanics

Generated by null/stub on May 9, 2026. Inputs: CWE entries + fix-commit diffs from this CVE's patches. Citations validated against bundle.

References

5

News mentions

0

No linked articles in our index yet.