A company is using the Generative AI Security Scoping Matrix to assess security responsibilities for its solutions. The company has identified four…

Building and training a generative AI model from scratch (Scope 4 in the matrix) assigns the maximum security ownership to the company. This scope requires the company to be responsible for the entire lifecycle, including the security of the training data, the integrity of the data processing and model training infrastructure, the security of the model architecture itself, and all post-training deployment and operational security. This contrasts with other scopes where a third-party vendor manages the security of the foundational model and its underlying infrastructure, creating a shared responsibility model.

1. AWS Documentation

Overview of Generative AI Security. This document introduces the Generative AI Security Scoping Matrix. It explicitly defines four scopes of use. Scope 4

"Build and train your own generative AI foundation model

" is described as the scenario where "you own all the security responsibilities." The accompanying diagram visually places this scope as having the highest level of customer responsibility.

2. Stanford University

CS329S: Machine Learning Systems Design

Lecture 15

MLOps: Tooling and Best Practices. This courseware discusses the end-to-end ML lifecycle. The lecture covers the extensive responsibilities involved in building a model from scratch

including data collection

infrastructure management

and deployment

which inherently include all associated security tasks

reinforcing that this path entails maximum ownership. (Reference to the principles of MLOps and full lifecycle ownership).