Is microsoft/Phi-4-multimodal-instruct safe to load?
CyberXYZ rates microsoft/Phi-4-multimodal-instruct LOW. No version-matched advisories and no malicious packages. It depends on 274 tracked packages, and 10 carry advisories. It ships Python that runs when the model is loaded (trust_remote_code).
MEDIUMconfig.json declares `auto_map` — from_pretrained(trust_remote_code=True) executes Python shipped in this repo.
2 more findings on this model.
- Author
- microsoft
- Task
- automatic-speech-recognition
- Loader
- transformers
- License
- mit
- Downloads
- 262,774
- Remote code
- ships Python that runs on load (trust_remote_code)
42 direct, 33 with findings, 256 reachable
| Package | Evidence | Advisories | Worst |
|---|---|---|---|
| pillow | direct, runtime | 147 | none pinned |
| transformers | direct, runtime | 24 | none pinned |
Is microsoft/Phi-4-multimodal-instruct safe to load?
CyberXYZ does not call a model safe in absolute terms. It rates microsoft/Phi-4-multimodal-instruct LOW. No version-matched advisories and no malicious packages. That covers its dependencies and how it loads, not every file it ships, so treat any check that did not run as unverified and check the exact files before you load them.
What does microsoft/Phi-4-multimodal-instruct depend on?
CyberXYZ tracks 274 dependencies for microsoft/Phi-4-multimodal-instruct, 42 of them direct. 10 carry advisories. They include pillow and transformers. A free account lists all of them.
Does microsoft/Phi-4-multimodal-instruct run code when it loads?
Yes, it can. microsoft/Phi-4-multimodal-instruct ships Python code and HuggingFace marks it trust_remote_code, so loading it with trust_remote_code enabled runs that code on your machine. Read the code first.
How do I check microsoft/Phi-4-multimodal-instruct before loading it?
Run the xyz MCP tool model_load_safety on microsoft/Phi-4-multimodal-instruct to get a verdict on its weight files before you load them, or search it in CyberXYZ Check at https://cyberxyz.io/check?q=microsoft%2FPhi-4-multimodal-instruct.
Risk computed 2026-11-14 at 08:51 UTC. Dependencies come from the model's requirements, its declared loader and the code it ships; advisories from NVD, GHSA and OSV.