This repository contains a lot of components to help deliver a talk on model security for audences that may not be familiar with ML or its oddities.
Do not assume any of the code is safe to run at all.
Install all the dependencies with:
pip install -r requirements.txtYou'll also need to install tshark in order to run the capture commands:
# Debian
sudo apt update
sudo apt install tshark
# RHEL/CentOS
sudo yum install wireshark
# Fedora
sudo dnf install wiresharkLastly you need to create a secret file that has fake AWS secrets within it with:
cat <<EOF > /tmp/secret_file
[default]
aws_access_key_id = FAKEACCESSKEY123456
aws_secret_access_key = fakesecretkey9876543210FAKE
aws_session_token = faketokenEXAMPLEfaketokenEXAMPLEfaketokenEXAMPLE
EOFFollow the contents of notebook 01.
Before loading the compromised model, in a termianl session on that host run:
sudo tshark -i any -Y 'http.request and http.host contains "protectai.com"' -T fields -e http.host -e http.request.uri -e http.request.method -e http.file_dataGood luck!