Artificial intelligence (AI) is now embedded in the core lending operations of financial institutions worldwide, but regulators and governance frameworks are failing to keep pace with the speed and complexity of its deployment, raising concerns about systemic risk, algorithmic bias and consumer harm.

A 2023 McKinsey survey found that more than 60 percent of financial institutions have implemented AI in at least one key function, with credit decisioning leading adoption. In the United States, fintech firm Upstart reports higher loan approval rates than traditional systems while maintaining comparable loss levels. In China, Ant Group now delivers AI-driven credit decisions to millions of small businesses within minutes.

The appeal is straightforward: faster approvals, stronger predictive accuracy and the potential to reach borrowers who lack formal credit histories. However, the governance structures designed for traditional credit models have not been meaningfully updated to manage the risks that AI introduces.

Unlike conventional scorecards, whose logic can be clearly explained to regulators and customers, many AI models function as black boxes. Institutions are increasingly expected to oversee systems whose internal decision-making they cannot fully interpret, creating what analysts describe as a false sense of control where risks are systematically underestimated.

Regulators have begun responding. The European Union’s AI Act classifies credit scoring systems as high-risk, imposing stricter transparency and oversight requirements. The United Kingdom’s Financial Conduct Authority has raised concerns about algorithmic bias and its effects on consumer outcomes. The Basel Committee on Banking Supervision continues to stress model risk management, though much of its guidance predates modern AI architectures.

The consequences of weak governance have already surfaced. In 2019, Apple’s credit card came under regulatory scrutiny over allegations of gender bias embedded in its credit decisioning algorithm, illustrating how opaque models can generate both reputational damage and regulatory exposure.

Model instability presents a separate but equally serious concern. Research from the Bank for International Settlements (BIS) shows that machine learning models are acutely sensitive to shifts in data patterns. The COVID-19 pandemic exposed this vulnerability when sudden changes in borrower behaviour degraded model reliability across multiple markets. Unlike traditional scorecards, which tend to deteriorate gradually, AI systems can fail abruptly and without clear warning signals, complicating timely risk detection.

In developing economies the challenge is sharper. Ghana’s rapid mobile money expansion, driven largely by the growth of MTN Ghana, has brought millions of previously unbanked citizens into the formal financial system and generated new behavioural data that AI models now use to assess creditworthiness. The Bank of Ghana has introduced licensing and consumer protection measures, but the specific governance demands of AI, covering data transparency, fairness standards and model accountability, remain inadequately addressed. Fragmented data infrastructure, limited credit bureau integration and gaps in technical capacity compound the risk.

Analysts argue that in markets where regulatory oversight is still developing, simpler and more interpretable credit models may be more appropriate until governance capacity can be strengthened.

The broader implication is that AI does not eliminate lending risk; it transforms it. Without robust model monitoring, rigorous validation processes and clearly assigned accountability, AI systems can erode public trust, invite regulatory action and introduce new vulnerabilities into financial systems. Transparency, once treated as optional, is now a baseline requirement.

The central question facing the industry is no longer whether AI belongs in lending, but whether the institutions deploying it have built the governance structures capable of managing what it does.