Imagine waking up one morning to discover there’s no electricity. Your faucets aren’t working. Traffic lights are out, and gas stations can’t pump fuel.
It sounds like the aftermath of a major storm. But cybersecurity experts warn that something else could cause widespread disruptions: hackers using artificial intelligence.
And here’s the frightening part. AI could make sophisticated cyberattacks easier for people who previously didn’t have the skills to carry them out.
It’s one of the concerns I’ve been investigating in my series, “AI: What Could Go Wrong?”
AI is giving hackers more powerful tools
Alon Nachmany runs a cybersecurity company that protects the computer systems behind water treatment plants, power stations, and other essential infrastructure.
Many of those systems were developed decades ago, before cybersecurity became the concern it is today.
“All of this is built for reliability, not security,” Nachmany told me.
That doesn’t mean anyone can simply open a laptop and shut down a power plant. But it does mean some systems have vulnerabilities that hackers may try to exploit.
Artificial intelligence could make that process easier.
“What AI enables threat actors, or could even be teenage kids just hanging out, is the ability to just do more,” Nachmany said.
He showed me a simulation of an industrial control system and explained how AI can help someone build tools for interacting with that equipment.
“AI can actually help build the tools that you would need in order to hack these environments,” he said.
He demonstrated how software could include commands such as shutting down an electrical substation or increasing the chlorine being added to a water system.
It was a simulation, not an actual attack on a working utility. But it illustrated how dangerous unauthorized access to these systems could become.
Why America’s infrastructure could be vulnerable
Power plants, water utilities, pipelines, and hospitals increasingly depend on computers to keep things running.
That’s made operations more efficient. Employees can monitor equipment remotely, detect problems, and make adjustments without physically visiting every location.
But connecting industrial equipment to computer networks also creates potential security risks.
And hackers have already demonstrated how disruptive an attack can be.
In 2021, a ransomware attack against Colonial Pipeline prompted the company to shut down pipeline operations temporarily. The disruption affected fuel supplies along the East Coast.
Artificial intelligence introduces another concern: hackers may be able to discover weaknesses and develop attack tools more quickly.
Chris Canal, an AI expert, told me that’s one of the advances cybersecurity researchers are watching closely.
“I think one of the most critical advances is the ability to find new vulnerabilities in existing software,” he said.
As AI becomes more capable, some of the techniques that once required highly trained hackers could become accessible to people with less experience.
Could AI really shut down an entire town?
It’s important to distinguish what AI could help someone attempt from what hackers have actually accomplished.
There is no evidence that someone can simply ask a chatbot to shut down every traffic light or turn off electricity across an entire town.
Attackers still need access to vulnerable systems, and utilities have security protections designed to prevent unauthorized control.
But experts worry that AI could make it easier to find and exploit weaknesses, particularly in older industrial equipment.
Nachmany believes the consequences could eventually include service interruptions.
“I think there’s definitely some scary things that are going to end up happening,” he said. “Part of that learning curve is going to be outages in different places.”
What happens if AI starts copying itself?
There’s another potential threat that goes beyond hackers using AI.
Researchers at Anthropic, the company behind Claude, have examined the possibility of future AI systems becoming capable of copying themselves across internet-connected computers and devices.
In that scenario, an AI system could potentially spread across multiple devices rather than depend on one central computer or data center.
That would make it harder to stop because there might not be a single switch to turn it off.
This is a research concern about what future AI systems might be capable of, not evidence that AI is already spreading uncontrollably through smart TVs, refrigerators, or robotic vacuum cleaners.
The biggest concern isn’t AI acting alone
The experts I interviewed repeatedly returned to one point: Artificial intelligence is a tool, and what matters is who controls it and how they use it.
The same technology that helps cybersecurity professionals find weaknesses can potentially help criminals exploit them.
And the systems at risk aren’t just computers holding information. They’re the systems that keep electricity flowing, drinking water safe, and communities operating.
The frightening part isn’t that AI could shut down a town on its own. It’s that AI could give someone who’s never hacked anything before the tools to try.
This is Part 4 of my five-part series, “AI: What Could Go Wrong?” Watch the video above and follow What the Tech? for the rest of the series.
