Operationalizing AI Governance Requirements through Rule-Based Mechanisms in Digital Infrastructures

Sammanfattning

Artificial intelligence (AI) systems are increasingly used to support decision-making inorganizational contexts, including areas such as recruitment, healthcare, finance, and publicservices. As these systems become embedded within digital infrastructures, questions oftransparency, accountability, fairness, human oversight, and risk management becomeincreasingly important. Although regulations, standards, and academic literature providegrowing guidance on responsible AI governance, a central challenge remains: governancerequirements are often defined at a high level and are not clearly translated into mechanismsthat can be implemented within technical and organizational processes. This thesis investigates how AI governance requirements can be operationalized and embeddedwithin digital infrastructures using rule-based mechanisms to support responsible AIdeployment. The study adopts a Design Science Research approach and develops a conceptualframework as its main artifact. The framework connects four components: governancerequirements, rule-based mechanisms, AI lifecycle stages, and digital infrastructurecomponents. Governance requirements are derived from regulation, standards, and academicliterature, including the Artificial Intelligence Act and the AI Risk Management Framework.These requirements are then translated into rule-based mechanisms expressed as explicitconditions and actions, mapped to governance-oriented AI lifecycle stages, and embeddedwithin digital infrastructure components such as data pipelines, model validation processes,decision interfaces, and monitoring systems. The framework is demonstrated through a literature-grounded scenario based on an AIrecruitment system used for candidate screening, ranking, and decision support. The scenarioillustrates how requirements related to fairness, transparency, human oversight, documentation,and monitoring can be translated into concrete rules and connected to specific systemcomponents. The demonstration shows that the proposed framework provides a structured wayto move from high-level governance requirements to system-level implementation. The thesis contributes to AI governance research by offering a mechanism-oriented perspectiveon operationalization and by connecting AI governance literature with digital infrastructuretheory. Practically, it provides a structured approach that can support organizations and systemdesigners in embedding governance requirements into AI systems. While the framework isconceptual and evaluated through a scenario rather than a real-world implementation, itprovides a foundation for future empirical validation and technical development.