Securing generative AI: Applying relevant security controls

This article discusses security controls to apply when securing a generative AI application, based on the scope of the application as defined in the Generative AI Security Scoping Matrix introduced in a previous article.

Specifically, the article covers:

• Controls for Scope 1 (consumer applications) and Scope 2 (enterprise applications)
• Controls for Scope 3 (using pre-trained models)
• Additional controls for Scope 4 (fine-tuning models)
• Considerations for Scope 5 (self-training models from scratch)
• Mapping of controls to MITRE ATLAS mitigations