Securing AI/ML Pipelines in the State Government Sector: Threat Models, Controls, and Governance

Government authorities and public sector organizations increasingly deploy artificial intelligence and machine learning systems in critical services including benefits eligibility verification, fraud detection, visitor flow optimization, and health monitoring. However, existing cybersecurity frameworks inadequately address vulnerabilities inherent in machine learning pipelines. Traditional security controls protect infrastructure and data storage against conventional attacks but fail to counter AI-specific threats such as data poisoning, model extraction, and adversarial manipulation, which require specialized protective mechanisms. A significant gap exists between the rapid adoption of AI technologies and the maturity of security governance frameworks in public sector environments. This article presents a comprehensive threat model addressing vulnerabilities across the entire machine learning lifecycle. Technical controls including data provenance verification, cryptographic model signing, secure training environments, and continuous inference monitoring form the foundation of a secure MLOps architecture. Alignment with the NIST AI Risk Management Framework provides structured approaches for risk identification, assessment, and mitigation. Clear governance accountability structures are established through system owners, independent risk reviewers, and oversight committees. Fairness considerations integrate alongside security controls to address public sector equity obligations. Audit trail requirements satisfy transparency mandates and support retrospective review of algorithmic decisions. This framework enables innovation in artificial intelligence while maintaining compliance requirements essential for State government applications.