Toward an MLSecOps 2.0 framework
Shifting MLSecOps to be intentionally integrated into AI solutions’ life cycle requires understanding and mapping its stages. AI life cycles will vary, depending on the model and data source (own or external), model hosting (own or third-party), the solution packaging (backend API, web application, or mobile app), and the type of AI (predictive or generative). Nevertheless, we can identify distinct steps commonly used in these variations. We can use them to create reusable patterns of MLSecOps orchestration. The patterns highlight the applicable security controls and the role of CI pipelines, MLOps, and SCM.
A crucial element is the orchestrator of flows across the AI solution spectrum, offering a single control pane ideally.
Let’s explore the options for this crucial role.
MLSecOps orchestration options
CI pipelines are an ideal candidate to act as the orchestrator of MLSecOps flows, and there are good reasons for that:
...
