Enhanced FoodieAI threat model
Based on the architecture, we can identify the following aspects:
- Trust boundaries: In our example scenario, the systems that are outside our trust boundary are social login authentication, ChatGPT LLM, Hugging Face, TensorFlow Hub, the external images database we use for ingredient recognition, and the sales and marketing system, which provides consolidated ingredient data, including pricing and user reviews. These are systems whose security is someone else’s responsibility, and we need to secure the communication with our system.
To reduce the complexity of our model, we also assume that the development of the mobile app is another team’s responsibility, which will apply mobile app security and is therefore outside our trust boundary.
- Critical assets and data sensitivity: Our system does not contain personal data. It uses sensitive commercial data in the ingredients database, fine-tuning dataset, and embedding derivatives. Furthermore...
