Historically, lenders failed differently. In the future, they may fail together

For decades, financial institutions invested heavily to build differentiated credit capabilities. Different underwriting philosophies. Different portfolio strategies. Different risk appetites. Different ways of understanding borrowers. Those differences mattered. Because they ensured that institutions did not all make the same mistakes at the same time. Today, I wonder if we are quietly moving in the opposite direction. The 2008 financial crisis was partly accelerated by a different kind of concentration. Nearly every major bank used similar Value at Risk models. When those models reached the same conclusions under stress, the consequences were amplified across the entire system simultaneously. The FSB has now named the equivalent risk for the AI era. In its 2025 report on AI in financial services, the Financial Stability Board warned explicitly: "The widespread use of common AI models and data sources could lead to increased correlations in trading, lending, and pricing — amplifying market stress and exacerbating liquidity crunches." Researchers have given this a name: algorithmic coupling. When multiple institutions deploy similar AI architectures — trained on comparable datasets, optimised for similar objectives, subject to similar failure modes — their decisioning becomes correlated even without any direct financial linkage between them. The efficiency gains are undeniable. The concentration risk is less discussed. Consider what algorithmic coupling looks like in practice during a macroeconomic shock: → A borrower segment is simultaneously downgraded across multiple lenders → Credit lines are reduced at the same moment across the system → Approval rates decline in coordinated fashion → Collections strategies tighten in parallel Not because institutions independently arrived at the same view through independent judgment. But because they are powered by similar decisioning infrastructure — responding to similar signals, crossing similar thresholds, at the same time. This is not a model risk issue. It is a systemic risk issue. The question is not whether an individual model is accurate. The question is what happens when a large part of the financial system reaches the same conclusion at the same moment. I have spent two decades building credit risk infrastructure across India, the UK, and Southeast Asia. The institutions that weathered cycles best were not those with the most sophisticated models. They were those that preserved independent risk judgment alongside their models — where governance actively challenged outputs rather than simply consuming them. Only 12% of banks today describe their AI governance framework as highly developed. Yet over 70% are already deploying AI in credit and risk functions. That gap is where the next systemic event may quietly be building. The strongest institutions will continue to embrace AI. But they will also ask harder questions: → Are we becoming overly dependent on external decisioning ecosystems? → Do we understand how our models behave under correlated stress — not individual stress? → Are we preserving independent risk judgment or outsourcing it? → Have we created sufficient diversity in our decision-making architecture? Technology concentration is already a recognised risk in cloud infrastructure, payments, and market utilities. The FSB, BIS, and Basel Committee are all watching it in AI. Credit decisioning may be next to attract that scrutiny. Because resilience is not created when everyone uses AI. It is created when everyone does not think exactly alike.

To CROs, CDOs, and Board Risk Committees — particularly those overseeing institutions operating across multiple geographies: Has your governance framework been updated to address algorithmic coupling — or is it still designed only for individual model validation? #CreditRisk #AIGovernance #ModelRisk #DigitalLending #Fintech #CDO #SystemicRisk #RiskLeadership