Banks Could Be Forced to Rein In Dangerous AI. Here's Why

Jan de Vries · 2026-08-13

It's one of those headlines that makes you do a double take: banks could actually face bans on using certain types of AI. That's not just a hypothetical scenario cooked up by tech skeptics. A new report from Durham University is pushing regulators to get serious about the risks lurking inside the financial system's growing reliance on artificial intelligence. For years, we've been told that AI is the future of banking—faster fraud detection, smarter lending decisions, better customer service. And sure, there's a lot of truth to that. But the same technology that helps a bank spot a suspicious transaction in milliseconds can also make decisions that are impossible to explain, let alone challenge. That's where the danger creeps in. ### The Problem With Black Box Decisions Here's the thing about modern AI systems: they're often so complex that even the engineers who built them can't fully trace how they arrive at a specific conclusion. In banking, that's a massive red flag. If a loan application gets denied or a credit limit gets slashed, there needs to be a clear, understandable reason. When the logic is buried inside a neural network, you're left with a "trust us, it's the algorithm" answer. That's not good enough anymore. Durham University's research points to a growing list of concerns that go beyond just customer frustration. Think about systemic risk. If every major bank is using similar AI models trained on similar data, they could all make the same bad call at the same time. That's not just a glitch—that's a potential financial crisis waiting to happen. ### What the Research Actually Recommends The Durham team isn't saying we should throw the whole concept of AI in finance out the window. Instead, they're calling for a more nuanced approach. Here's what they want regulators to seriously consider: - **Outright bans on the most dangerous applications**—specifically, AI systems that make irreversible decisions with no human oversight. - **Mandatory transparency requirements** so that banks have to prove they understand how their AI models work. - **Stress testing for AI** that's similar to the stress tests banks already run for their capital reserves. - **Accountability structures** that assign clear responsibility when an AI-driven decision causes harm. The last point is a big one. Right now, if an AI system makes a mistake that costs someone their home or their savings, who do you blame? The software? The vendor? The bank executive who signed off on it? The researchers argue that we need clear rules on liability before something goes catastrophically wrong. ### Why This Matters Beyond the Banking Sector You might be thinking, "I don't work in banking, why should I care?" Well, because the financial system touches everything. If you've ever applied for a mortgage, opened a credit card, or even just used a mobile payment app, AI is already involved in your financial life. And the safeguards that get put in place for banks often become the template for other industries down the road. There's also a competitive angle to consider. European banks have been pushing hard on AI innovation, but they've also been more cautious about regulation than their American counterparts. If the EU moves forward with stricter rules, it could change the global playing field. Banks operating in the United States might find themselves under pressure to adopt similar standards just to stay competitive in international markets. The conversation is just getting started, but the direction is clear. The era of "move fast and break things" in finance is coming to an end. The next few years will likely determine whether AI becomes a trusted partner in banking or a liability that needs to be kept on a very short leash.