Instructions to use Kernelwraith/keras-native-zip-expansion-poc-20260521 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Keras
How to use Kernelwraith/keras-native-zip-expansion-poc-20260521 with Keras:
# Available backend options are: "jax", "torch", "tensorflow". import os os.environ["KERAS_BACKEND"] = "jax" import keras model = keras.saving.load_model("hf://Kernelwraith/keras-native-zip-expansion-poc-20260521") - Notebooks
- Google Colab
- Kaggle
Keras Native .keras ZIP Expansion PoC
This repository is a gated proof-of-concept for a security report submitted through Huntr's model file format bounty process.
Purpose
The PoC demonstrates that Keras Native .keras model loading can extract a compressed, unrequested archive asset to disk while loading with safe_mode=True.
Tested Target
- Project:
keras-team/keras - Version observed by PoC:
3.15.0 - Commit:
0172a2f5e42e227c120c65da5daef3ebbfaaf06d - Backend used for the local proof:
numpy
Files
keras_native_zip_expansion_poc.keras: generated local PoC model file.poc_keras_native_zip_expansion_dos.py: local-only reproduction script.local_reproduction_output_20260521.json: proof output from a local run.
Local Reproduction
KERAS_BACKEND=numpy python poc_keras_native_zip_expansion_dos.py --payload-mib 64
Expected markers:
"safe_mode": true
"added_asset_uncompressed_size_bytes": 67108864
"relative_path": "assets/unrequested_padding.bin"
"finding": "REPRODUCED_KERAS_NATIVE_LOAD_MODEL_EXTRACTS_UNREQUESTED_COMPRESSED_ASSET"
Boundary
This is a local-only proof. It does not contact Keras, Huntr, Hugging Face services during execution, cloud services, production systems, or external targets.
Access should remain gated and granted only to the reviewing party requested by Huntr.
- Downloads last month
- -
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support