Concentration, aggregation and where the risk lands

The Monoculture Problem

Every failure we've looked at so far has been one organization's problem. This lesson is about the one that isn't. Think about monoculture in farming. You plant the same variety of wheat across every field because it's the most efficient. Every field is productive, but every field is also vulnerable to exactly the same disease on exactly the same day. Diversity is inefficient right up until the moment it's the only thing that saves you. In the AI world, an organization running a product built on a model it's never heard of is like that farmer. They don't realize they've planted the same 'crop' as everyone else until the blight hits.

The Hidden Foundation Layer

If you look at what sits underneath all this AI, there's a great deal less variety than the number of products would suggest. As we touched on in our look at third-party AI, there's a small number of foundation models underneath a very large number of products. While the software on the surface looks different, the 'engine' underneath is often identical. This creates a shared dependency that most users haven't inventoried and simply cannot see.

Model Uniformity

In traditional human processes, errors are usually independent. If ten different people summarize a report, they might make ten different mistakes that eventually cancel each other out. AI introduces model uniformity. Because many organizations use similar models in similar ways, their errors align and compound rather than offsetting. It's not just that the AI failed; it's that it failed for everyone at once.

Takeaways: Concentration & Uniformity

Let's recap. We've seen how a small number of models sit underneath a vast number of products, creating hidden, shared dependencies. We've also defined model uniformity, where AI errors align rather than canceling out, turning independent risks into correlated ones.

Correlation: The Insurer's Fear

Correlation is different in kind from any individual failure. Independent risks are manageable through the law of large numbers. Correlated risks—where one event triggers many losses simultaneously—are not. This is why AI risk is a systemic concern. According to a SUERF (2026) policy note, there are five features by which AI can amplify this systemic financial risk: concentration, uniformity, monitoring difficulty, over-reliance, and speed.

Mapping the Exposure

To manage this, you have to locate the exposure. It usually sits in one of four places. Click on each area to see how a single failure often lands in more than one box. The insured carries risk in their own operations. Your own firm is exposed through internal tools. Third parties like vendors are the source. And finally, the portfolio accumulates the silent risk across many clients.

Takeaways: Correlation & Location

Let's summarize correlation and location. We've learned why correlation is the insurer's defining fear and how risk lands in multiple places at once. Use these definitions to distinguish between individual failures and the systemic aggregation we've discussed today.

The Industry Landscape

We'll end on an honest note: a market-wide AI correlated loss hasn't happened yet. But the industry is preparing. As evidenced by the Lloyd's Market Association (2025) survey on AI loss scenarios, the market is actively trying to map these systemic events. Measuring accumulation, and doing something about any of this, is the work of the next two courses. That is the landscape. You can now name what goes wrong with these systems and say where the exposure sits, which is more than most people in this market can do.