Human-in-the-Loop (HITL) in AI-Assisted Disclosure

Artificial Intelligence (AI) has become an integral part of corporate disclosure processes, changing the way that companies create, validate and communicate information to others. The use of AI in decision making and disclosure processes, has revolutionized the governance, in the legal, financial and public administration sectors. Its critical, especially in high-stakes settings such as financial reporting and legal disclosure requirements. Although AI-assisted disclosure is advantageous in terms of efficiency, scalability and analytical capabilities, it poses governance, auditability and explainability challenges. These dimensions are interdependent and current study explores these dimensions highlighting the need for accountability and compliance of human-in-the-loop (HITL) supervision. The study suggests an integrated architecture which includes governance structures, audit trails, explainability techniques and structured human monitoring. Current research analyzes disclosure processes and introduces a new AI-assisted disclosure approach with a Human-in-the-Loop (HITL) system that allows for human oversight, accountability, and trust. EU AI Act, NIST AI RMF, and emerging compliance and open governance practices, should be reviewed regarding their application to HITL systems to investigate trends and patterns in managing ‘black box' risks, audit trails and meaningful explanations. Technical (e.g., XAI techniques such as SHAP/LIME) and procedural safeguards must be combined to provide an effective governance process, holistic HITL supplemented by technical safeguards do not rubber stamp decisions, but instead increase decision quality. Good governance needs to be multi-layered in which there is structured human intervention, regulatory control and technical transparency. A full package approach, combining audit trails, explainability protocols and human review processes is recommended to ensure legally compliant and ethically responsible disclosure when made by AI.