- Describe the architecture of your application - include where user (or attacker) input comes in, where it is rendered, any storage used by the LLM (e.g. for RAG), training data, fine-tuning data, ‘memory’ or caching mechanisms and any external data sources
- Walk through the flow-chart to understand risk factors, outputs and attack vectors - add them to your diagram as you go
- Add each ‘risk’ to your diagram and understand the applicability of each one
- Apply your defences to ensure you are mitigating all risks
{% @mailchimp/mailchimpSubscribe %}