Consider the following activities performed during the Monitoring and Maintenance stage of the AI security life cycle:I. Identifying Model Drift via…

The Monitoring and Maintenance phase of the AI security life cycle involves activities performed after a model is deployed. This includes continuously monitoring for performance degradation like model drift (I), performing scheduled retraining to update the model and maintain its accuracy (II), and patching the underlying infrastructure, such as virtual containers, to protect against new vulnerabilities (III). Data sanitization on the initial training set (IV) is a data preparation activity that occurs during the development phase, before the model is deployed and monitored.

A. This option is incomplete as it omits the critical security maintenance task of patching the operational environment (III), which is essential for ongoing security.

B. This option is incorrect because data sanitization (IV) is a pre-deployment activity performed during the data preparation stage of the AI development life cycle.

D. This option incorrectly includes data sanitization (IV), which is part of the initial data preparation and development phase, not the post-deployment monitoring and maintenance phase.

1. National Institute of Standards and Technology (NIST). (2023). AI Risk Management Framework (AI RMF 1.0). The "Govern" function is mapped across the lifecycle, and the "Operate & Monitor" stage explicitly covers post-deployment activities, including performance monitoring (related to I) and managing system security (related to III).

2. Microsoft. (2020). Machine learning operations (MLOps) framework. Azure Architecture Center. The "Model deployment & monitoring" phase of the MLOps cycle includes monitoring for drift (I) and triggering retraining (II). The overall operational management includes infrastructure security (III).

3. Kumar, S., et al. (2022). Engineering Trustworthy AI Systems: A Development and Operations (DevOps) Perspective. arXiv:2206.00535. Section 4.3, "Monitoring and Maintenance," details the need for continuous monitoring for concept drift (I), model retraining (II), and infrastructure security management (III).