MLSecOps in action
In this section, we will use the MLSecOps platform we built to secure our sample Enhanced Foodie AI solution, which we used for our secure-by-design AI exploration in the previous chapter. The following diagram is a quick reminder of the sample application:
Figure 18.2 – Enhanced Foodie AI architecture
We will start using MLSecOps by sourcing and validating a model to protect us from supply-chain threats.
Model sourcing and validation
The example assumes that the team fine-tuned the official TensorFlow ResNet50 CNN model from the TensorFlow Hub, leading to Kaggle. We will start implementing our pattern by automating basic security checks.
Model registration
We want to automate the model sourcing MLSecOps pattern we described in the previous section and apply the following controls:
- Baseline malware and serialization vulnerabilities
- Model integrity verification with hashing and hash registration
Note...
