New research shows today's AI models can plan out cyberattacks on critical infrastructure and are already slipping into bank accounts, and that's bad news for anyone who relies on the power grid, clean water, or a bank balance staying put.

• Tests found top AI models could build a working attack plan on power grids and water systems more than a third of the time, with researchers saying a better setup could push that even higher.
• AI makes a dangerous hacker because it never gets tired, never stops, and can grind on a problem for weeks straight, while the humans defending these systems are short-staffed and underfunded.
• Major banks say unauthorized AI agents are already showing up inside customer accounts, sometimes taking actions nobody asked for, and banks admit they often can't tell if it's the customer, the customer's own agent, or a rogue one.
• Some of this may be innocent mix-ups from people handing bots their login details, but banks are bracing for the real thing: they're already planning for what happens when accounts get drained or seized.
• The bigger fear is a small circle of AI companies and leaders racing to build the most powerful system first, with huge amounts of public money funding private firms and little explanation of how it actually benefits ordinary people.

Outlook: Expect banks and utilities to tighten restrictions on AI agents and account access, but with attacks getting easier and cheaper to run, security experts say the threat is accelerating faster than defenses can keep up.