Security protects data, models, and infrastructure across the ML lifecycle. In production, weak data security can expose sensitive information, violate privacy obligations, and create financial or legal risk; deployed models must also resist information leakage and adversarial attacks. section 2.2 section 10.1
Security concerns appear at several operational boundaries:
- Data pipelines require privacy protection and robust security while preserving visibility at scale. section 2.2
- Model-serving environments must prevent sensitive information leakage and protect deployed models from adversarial attacks. section 10.1
- Security decisions interact with deployment reliability, model versioning, resource management, and explainability requirements. section 10.1
