The adoption of artificial intelligence is rapidly transforming how financial institutions manage compliance and financial crime risks, with machine learning systems now monitoring transactions, identifying anomalous patterns, and automating regulatory reporting across global financial networks. While these technologies promise greater operational efficiency and advanced control capabilities against illicit financial flows, their integration into compliance infrastructures presents a governance dilemma, as automated systems operating within punitive regulatory contexts can reinforce defensive compliance behavior and inadvertently contribute to financial exclusion. Over the past two decades, global regulatory frameworks against money laundering and terrorist financing have expanded significantly, with standards developed by the Financial Action Task Force (FATF) strengthening requirements on customer due diligence, beneficial ownership transparency, and transaction monitoring, while enforcement actions and sanctions have become more frequent. In this context, banks must continuously assess regulatory risk associated with customers, economic sectors, and geographical areas, with AI playing an increasingly central role by analysing large volumes of financial data to identify suspicious patterns and flag transactions for scrutiny. However, because the cost of failing to detect suspicious activity can be very high, algorithmic monitoring systems tend to prioritise reducing false negatives, leading risk classification systems to adopt more conservative assessments that may classify even fully legitimate customers or transactions as high-risk. Consequently, financial institutions often adopt increasingly prudent compliance strategies, frequently interpreting regulatory obligations defensively when expectations are ambiguous or potential penalties are very high.