Downloads · 30 days
2
5% of all-time downloads
celvexgroup/modelaudit-tflite-builtincode-differential-poc
modelaudit-tflite-builtincode-differential-poc is a machine learning model from celvexgroup. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
Coordinated disclosure PoC for huntr (Protect AI MFV). Benign payload only (no custom kernel shipped).
Downloads · 30 days
2
5% of all-time downloads
All-time downloads
38
Public
Repo size
956 B
Likes
0
Public
Click a slice to open those files.
Other1.5 KB · 42%
From the Hugging Face model README
Coordinated disclosure PoC for huntr (Protect AI MFV). Benign payload only (no custom kernel shipped).
modelaudit==0.2.49 with the modelaudit[tflite] extra (tflite bindings 2.18.0), Python 3.11.15.modelaudit scan bypass.tflite -> Clean / exit 0, but every conformant TFLite/LiteRT runtime
resolves the op to CUSTOM (evil_marker_op). The control_*.tflite files are the SAME model with the fields set so
modelaudit correctly fires CRITICAL S302 (exit 1) - proving the detector is live and bypass is a real miss.BuiltinCode(), which returns deprecated_builtin_code when
builtin_code < 127; runtimes use max(builtin_code, deprecated_builtin_code). Setting builtin_code=CUSTOM(32),
deprecated_builtin_code=ADD(0) makes modelaudit see ADD and the runtime see CUSTOM.Files: bypass.tflite (scan Clean), control_new.tflite / benign_add.tflite (controls). Full report in the huntr submission.