The world's most important banking regulator — the Bank for International Settlements — is raising an alarm. After two years of studying artificial intelligence, it says the financial system faces real dangers from how AI companies are valued and how banks are using the technology.

The concern isn't that AI itself is bad. It's that too much money has piled into a handful of giant tech companies, and that banks are using AI without proper safeguards. If either of those things goes wrong, it could ripple through the financial system fast.

Money Is Concentrated in Too Few Companies

Most of the money flowing into AI is going to a few massive, well-established companies like Nvidia — not to startups or lesser-known names. On the surface, that sounds safer. These companies actually make profits, unlike dot-com bubble companies that burned through cash with no revenue. But concentration creates its own danger.

Imagine a crowd of people all standing on the same few floorboards of an old house. Those floorboards are fine right now. But if even one starts to weaken — because a company disappoints investors, or a regulator cracks down, or a competitor emerges — a lot of people are standing very close to the problem. In early 2025, the arrival of DeepSeek, a Chinese AI model, shook confidence that the current leaders really need as much computing power and spending as everyone thought. Prices dipped and recovered. But the underlying question remains: Are these valuations too high?

Banks Are Using AI Without Proper Rules

Here's the sharper worry. Banks are already using AI to make lending decisions, detect fraud, and trade. But they're often doing this without strong rules about how the systems work or what happens if they fail.

The problem is speed and hidden complexity. When banks used simpler tools, regulators could check the logic. With modern AI, even the people building the systems sometimes can't fully explain why the software made a particular decision. And if multiple banks use similar AI systems, they're all making similar mistakes at the same time — a so-called "correlated failure."

In October 2025, global financial regulators said they were stepping up monitoring of how AI is being used. The catch: regulators have limited staff, and the question is whether they can write enforceable rules fast enough, before problems compound.

Two Different Time Scales

Central banks face an odd puzzle. The big, long-term benefits of AI — higher productivity, stronger economic growth — will take years to clearly show up in the data. But the financial system is already exposed to AI in ways that need oversight right now: in stock portfolios, in lending models, in the systems that keep major banks running.

The BIS is signaling that the faster-moving risks matter most for the moment. Whether existing rules can adapt quickly enough is the question no one can yet answer.