Deepfake

Challenge Information

• Name: Deepfake
• Points: 100
• Category: AI
• Objective: Analyze the provided .h5 file to retrieve the flag.

Solution

1. Initial Investigation:

   • The challenge provided a corrupted HDF5 (.h5) file
   • Attempts to load the file using deep learning frameworks like TensorFlow or PyTorch failed, as the file structure appeared corrupted

2. Static Analysis:

   • Used h5py, a library for inspecting and manipulating HDF5 files, to analyze the file:

     1 2 3 4 5 6

     import h5py with h5py.File("file.h5", "r") as h5file: print(h5file.keys()) for layer in h5file.keys(): print(layer, h5file[layer][:])

   • Extracted the number of layers and their corresponding data.

3. Interpreting the Data:

   • The data extracted from the layers did not directly reveal the flag.

   • Through iterative debugging and suggestions (thanks, ChatGPT!), realized that rounding the values in the data to the nearest integers, we found that it appeared to be ASCII values.

     

4. Flag Recovery:

   • After decoding these decimal representations of ASCII, the flag was clearly encoded in the data.

     

Flag

• Flag: SHCTF24{Hidd3n_in_Th3_Lay3rs}