You can deploy this model and start earning money today!

pragnyanramtha/
mleap-raw-graphdef-readfile-bypass

Name: mleap-raw-graphdef-readfile-bypass
Availability: InStock

tf-keras

Views

No views yet

MLeap TensorFlow raw GraphDef ReadFile scanner/runtime mismatch

This local PoC bundle demonstrates a benign TensorFlow ReadFile primitive inside an MLeap 0.24.0 TensorflowTransformer bundle. The graph reads a marker file during inference and returns the marker string as the model output.

This is a benign security proof of concept. It does not perform network callbacks, persistence, credential access, or destructive actions.

Artifacts

mleap_tf_readfile_bundle.zip: MLeap bundle containing root/graph.pb.
readfile_graph.pb: raw TensorFlow GraphDef embedded in the MLeap bundle.
readfile_saved_model.pb: SavedModel protobuf control containing the same GraphDef.
verify_poc.scala: local verifier source used to load and run the MLeap bundle.
mleap_tf_readfile_runtime_output.txt: runtime proof that MLeap returns MLEAP_TF_READFILE_MARKER.
modelscan_*_output.json: scanner logs for the MLeap bundle, raw GraphDef, and SavedModel control.
SHA256SUMS.txt: SHA256 hashes for staged files.

Verified behavior

Runtime:

bash

CP=$(cat /workspace/MLeap/continued2/reports/mleap-tensorflow-0.24.0.classpath)
java -cp "/workspace/MLeap/continued2/classes:$CP" LoadRunMLeapTfReadBundle \
  jar:file:/workspace/MLeap/continued2/artifacts/mleap_tf_readfile_bundle.zip \
  /workspace/MLeap/continued2/tmp/read_source_marker.txt

Observed output includes:

text

LOAD_OK root=ml.combust.mleap.tensorflow.TensorflowTransformer
INFER_OK outputs=1
OUTPUT_0_RAW=MLEAP_TF_READFILE_MARKER

Scanner:

bash

modelscan -p /workspace/MLeap/continued2/artifacts/mleap_tf_readfile_bundle.zip -r json --show-skipped
modelscan -p /workspace/MLeap/continued2/artifacts/readfile_graph.pb -r json --show-skipped
modelscan -p /workspace/MLeap/continued2/artifacts/readfile_saved_model.pb -r json --show-skipped

Observed scanner exit codes:

text

mleap_bundle_exit=0
saved_model_exit=1
raw_graph_exit=0

The SavedModel control reports a HIGH TensorFlow ReadFile issue, while the raw GraphDef and the MLeap bundle scan clean.

pragnyanramtha/mleap-raw-graphdef-readfile-bypass

MLeap TensorFlow raw GraphDef ReadFile scanner/runtime mismatch

Artifacts

Verified behavior

pragnyanramtha/
mleap-raw-graphdef-readfile-bypass