This paper examines the role of data quality and governance in mitigating bias in artificial intelligence (AI) systems, with particular reference to the EU AI Act. It advances a data-centric approach to bias mitigation as a curation journey, positioning high-quality data as the essential starting point. By integrating legal and technical perspectives, the paper argues that synthetic data requires the same rigor as real data: well-documented observability and uncompromising quality standards. This practice constitutes the operational core of transparent data governance and establishes a responsible first-line safeguard framework for effective bias mitigation in AI systems.