AI's Hidden Risks: How a Single Glitch Could Freeze Your Bank Account

Jan de Vries · 2026-11-02

Imagine waking up to find your bank account frozen. Or a hospital appointment canceled because of a software error. Or grocery shelves empty due to a supply chain glitch. These aren't scenes from a dystopian novel—they're the potential consequences of AI failures that experts are warning about. Vendan Ananda Kumararajah, a researcher focused on AI risks, has outlined four hypothetical AI crises that could hit Britain. And while they're set in the UK, the lessons apply just as much to the US and the rest of the world. Because AI is everywhere—from your bank to your doctor's office to the trucks that deliver your food. ### The Four Crises Kumararajah's scenarios are fictional, but they're based on real vulnerabilities. Here's what he imagines: - **Frozen bank accounts:** A bug in an AI-driven fraud detection system locks thousands of customers out of their accounts for days. No access to cash, no way to pay bills. Panic sets in. - **NHS errors:** An AI used to triage patients misreads symptoms, leading to delayed treatments and misdiagnoses. Vulnerable patients suffer the most. - **Food shortages:** An AI that manages supply chains makes a miscalculation, causing empty shelves in supermarkets. People rush to stock up, making the shortage worse. - **Vulnerable communities hit hardest:** In each case, the people with the least resources—the elderly, the poor, those without backup plans—are the ones who bear the brunt. These scenarios might sound far-fetched, but they're grounded in reality. AI systems are complex, and when they fail, they can fail in unexpected ways. Just look at recent incidents: automated trading algorithms causing flash crashes, or facial recognition systems wrongly identifying innocent people. ### Why AI Crises Are Different What makes AI failures so dangerous? It's the speed and scale. A human error might affect a few transactions; an AI error can affect millions in seconds. And because AI systems are often black boxes, it can take days to figure out what went wrong. > "The risk isn't that AI will become evil. It's that it will do exactly what we programmed it to do—but in a context we didn't anticipate." That's a quote worth remembering. AI doesn't need to be malicious to cause harm. A simple coding error or a flawed assumption can snowball into a crisis. ### What Can We Do? So how do we prevent these nightmares? Kumararajah suggests a few safeguards: - **Stress-test AI systems:** Just like bridges are tested for extreme weather, AI should be tested for extreme scenarios. Red teaming—where experts try to break the system—can uncover weaknesses. - **Keep humans in the loop:** For critical services like healthcare and banking, there should always be a human oversight. AI can assist, but not replace human judgment entirely. - **Plan for failure:** Assume things will go wrong. Have backup systems and manual overrides ready. And make sure the most vulnerable are protected. - **Transparency:** Companies and governments should be open about how AI is used and what happens when it fails. Hiding mistakes only makes things worse. ### The Bottom Line AI has enormous potential to improve our lives, but it's not risk-free. The scenarios above are hypothetical, but they're a wake-up call. As AI becomes more integrated into our daily lives, we need to be proactive about safety and ethics. Because the next crisis might not be a hypothetical—it might be just a glitch away. For entrepreneurs and businesses, this is a reminder: build responsibly. For everyone else, it's a reason to stay informed and demand accountability. After all, technology should serve us, not the other way around. What do you think? Are we prepared for an AI crisis? The conversation is just getting started.