The Rise of the AI Monoculture
In the world of finance, speed and accuracy are everything. AI delivers both, automating everything from credit scoring and fraud detection to stock trading and customer service. The problem is that many banks, instead of building unique systems from scratch,
are turning to a small number of specialised third-party vendors and cloud providers. This creates what experts call a "model monoculture.” While it seems like hundreds of different banks are using their own AI, many are actually running on similar underlying models, trained on overlapping datasets, and hosted on the same few cloud platforms. This convergence means that a flaw or blind spot in one model is likely present in many others, setting the stage for a synchronized failure.
What Is Correlated Failure?
Imagine a scenario where dozens of independent banks use AI to manage their loan portfolios. Now, imagine an unexpected economic event occurs—one that wasn't in the historical data used to train their models. If all the AI systems were built on similar logic, they could all react in the same faulty way simultaneously. For instance, they might all drastically underestimate risk, leading to a wave of bad loans, or conversely, all freeze lending at once, causing a credit crunch. This is correlated failure: a single trigger causing a cascade of identical errors across what are supposed to be independent institutions. The result is that a localized shock can be amplified into a system-wide crisis, a concern regulators like the U.S. Securities and Exchange Commission have flagged.
The Vendor and Data Bottleneck
This risk is driven by two main factors: vendor concentration and data similarity. A large portion of corporate AI runs on infrastructure provided by a handful of tech giants. Furthermore, many financial institutions license sophisticated AI models from the same few specialized fintech vendors. While this is efficient, it creates a dependency. An issue with a single vendor's model could affect all of its clients at once. Compounding this is the reliance on common datasets to train these models. When everyone trains their AI on the same public economic data or industry benchmarks, the models develop similar 'worldviews' and, consequently, similar blind spots. An AI model trained only on data from a stable economy may not know how to react during a sudden downturn, a weakness that is replicated across all systems using that data.
A New Kind of Systemic Risk
Before AI, the risk of many banks failing together was often tied to direct financial exposure, like one bank owing another money. This new risk is more subtle. It’s not about direct connections but about a shared, invisible vulnerability within their core decision-making logic. Regulators are increasingly worried because this 'herding' behaviour, driven by algorithms, can happen much faster than human-led decision-making. The U.S. Office of the Comptroller of the Currency (OCC) has highlighted that while AI can strengthen risk management, its widespread adoption also introduces new challenges related to compliance, credit, and operational risk. Studies have even found that higher AI investment at banks can correlate with greater operational losses, particularly from system failures and external fraud, if not managed properly.
Navigating the Future
The solution isn't to abandon AI, but to manage it more intelligently. Financial watchdogs and industry experts are calling for greater diversity in the models and data sources that banks use. This includes more rigorous testing of third-party models and developing 'kill switches' to disable faulty AI before they can cause widespread damage—a feature a surprising number of banks admit they lack. Regulators are also beginning to demand more transparency, wanting to know how a bank's AI models make decisions, a challenge when many operate as 'black boxes.' Ultimately, the goal is to move from a fragile AI monoculture to a resilient ecosystem where diversity in models and data acts as a buffer against systemic shocks, ensuring the efficiency of AI doesn't come at the cost of financial stability.
