Most conversations about artificial intelligence eventually end up in the same place.
Will it replace jobs? Can students use it to write papers? Can businesses eliminate employees? Will computers eventually become smarter than humans?
Those are legitimate questions.
But there's another AI story developing almost entirely behind the scenes, and it has much more to do with the physical world than chatbots.
AI is beginning to interact with the infrastructure that keeps modern life functioning.
Electricity. Communications. Transportation. Water. Industrial systems.
The U.S. Department of Energy now describes AI as becoming integral to how the energy sector plans, operates, monitors and defends critical infrastructure. At the same time, DOE is specifically developing programs to protect infrastructure from AI-enabled attacks and to secure AI systems that may themselves be involved in operating or defending energy systems.
That's considerably more consequential than whether a chatbot can write a convincing email.
The Real Issue Is Interdependence
Here's the part that's easy to miss.
America's critical systems don't operate independently.
CISA identifies 16 critical infrastructure sectors, but four—energy, communications, transportation, and water—are particularly fundamental because so many other systems depend upon them.
And they depend upon each other.
Consider communications.
Cell towers and network facilities need electricity. Some depend on backup generators when electricity fails. Those generators need fuel, which requires transportation.
Meanwhile, electric utilities depend on communications networks to monitor and control portions of the grid.
Water utilities use communications systems to monitor water quality, treatment processes and distribution.
Financial institutions need communications networks to process transactions.
Emergency services need them for dispatch, alerts and coordination.
CISA specifically warns that because infrastructure is highly interconnected, a disruption in one system can cascade into other critical systems. (CISA)
That's the real story.
The concern isn't necessarily one giant AI system suddenly shutting down America.
It's the possibility that increasingly automated systems become connected to an already complicated web of dependencies.
AI Can Make Decisions Faster. That's Both the Advantage and the Risk.
Modern infrastructure produces enormous amounts of data.
Human operators can't manually examine every sensor reading, network event, weather forecast, equipment status and security alert in real time.
AI offers an obvious advantage: speed.
DOE is already researching AI applications for grid planning, operations, reliability, resilience and cybersecurity. One current DOE initiative aims to use AI to dramatically accelerate decision-making in grid operations and security. (The Department of Energy's Energy.gov)
But speed introduces an interesting problem.
A human making a bad decision can cause damage.
An automated system making a bad decision can potentially make that decision much faster.
This isn't hypothetical enough for researchers to ignore.
In September 2026, DOE described work with Sandia National Laboratories on an AI system designed to help grid-security operators identify cyber threats. DOE reported that the system could locate threats with 95% accuracy in testing and reduce some data-engineering work from months to hours.
But the next phase of the project specifically includes studying AI “hallucinations.”
Why?
Because DOE notes that an incorrect AI output in critical infrastructure could cause operators to miss an actual cyberattack—or falsely identify an attack that isn't happening. (The Department of Energy's Energy.gov)
That distinction becomes extremely important when software isn't recommending a restaurant or summarizing a document.
It's helping humans make decisions about infrastructure.
AI Creates a Second Risk Most People Never Consider
There are actually two separate problems.
The first is using AI safely inside critical systems.
The second is attackers using AI against those systems.
DOE's current AI-FORTS program explicitly separates the challenge into three categories: protecting infrastructure from AI-enabled attacks, using AI to improve infrastructure security, and protecting the AI systems themselves.
AI can help defenders analyze enormous amounts of information.
But many of those same capabilities can potentially help attackers automate portions of reconnaissance, vulnerability discovery, social engineering or other cyber activity.
NIST makes a similar point: AI may provide defenders with new cybersecurity tools while simultaneously enhancing the capabilities of those targeting information technology and operational technology systems.
That's one reason NIST began developing a dedicated Trustworthy AI in Critical Infrastructure profile in 2026.
This isn't a warning that AI shouldn't be used.
Quite the opposite.
AI could ultimately make infrastructure considerably more resilient.
The point is that connecting powerful automated decision-making systems to physical infrastructure creates an entirely different category of consequences when something goes wrong.
Your Debit Card Is a Good Example
Imagine electricity disappears across part of a region.
Your first thought might be:
I'll simply drive somewhere that still has power.
But follow the dependencies.
Some cellular infrastructure can continue operating temporarily using backup power, but extended outages create additional problems.
Communications disruptions can affect electronic transactions.
Fuel stations need electricity to operate pumps and payment systems.
Backup generators supporting telecommunications and other facilities need fuel deliveries.
Transportation itself depends on functioning infrastructure.
Now imagine payment processing becomes unreliable at the same time.
You might have $10,000 sitting safely in your bank account and still temporarily struggle to buy $30 worth of gasoline.
Your money didn't disappear.
Your ability to access the system connecting your money to the fuel did.
That difference is fundamental to understanding modern preparedness.
The Grocery Store Isn't Really a Warehouse
The same principle applies to food.
Most people see a supermarket filled with thousands of products and unconsciously think of it as a giant warehouse.
It's better understood as the visible end of a constantly moving logistics network.
Products arrive through transportation systems. Warehouses rely on electricity and communications. Retailers rely on inventory databases. Refrigerated products depend on continuous power. Transactions depend on payment networks.
The store isn't an isolated resource.
It's a node in a network.
And networks behave differently from individual machines.
A machine can fail.
A network can experience cascading failures.
That's why infrastructure experts spend so much time studying dependencies rather than simply individual components.
Why the First 48 Hours Feel Different
This also explains why the beginning of a disruption can be particularly confusing for households.
It's often an information problem before it becomes a supply problem.
Is this a neighborhood outage?
A citywide outage?
Will electricity return in 30 minutes?
Tomorrow?
Is cellular service affected?
Are stores operating?
Should you drive somewhere else or conserve fuel?
Nobody knows yet.
That uncertainty changes behavior.
Some people immediately rush out to purchase supplies.
Others assume everything will return to normal shortly.
Neither reaction requires society to “collapse.”
All it takes is enough people responding simultaneously to incomplete information.
That's where personal preparedness becomes useful.
Not because you're predicting an AI catastrophe.
Because having basic redundancy means you don't have to make every decision during the most confusing hours of a disruption.
The Bigger AI Story Isn't Really About AI
There's one more surprising piece to this story.
AI isn't only becoming something infrastructure may depend upon.
AI itself requires enormous infrastructure.
DOE reported that data centers consumed approximately 4.4% of U.S. electricity in 2023 and projects that share could rise as high as 12% by 2028 as data centers, AI applications and other demand expand.
So the relationship works in both directions.
AI may increasingly help manage electricity.
Electricity is simultaneously required to operate AI.
That creates yet another dependency.
None of this means we're headed toward some Hollywood-style machine takeover.
The more realistic issue is much less dramatic—and arguably more useful to understand.
We are connecting increasingly powerful software to increasingly interconnected physical systems.
Those systems can provide extraordinary reliability and efficiency.
But complexity creates dependencies.
Dependencies create potential failure paths.
And automation can make both successful decisions and unsuccessful ones happen faster.
That's the AI risk worth paying attention to.
Not because everyone should be afraid of artificial intelligence.
But because preparedness begins with understanding what you're actually dependent upon.
And today, turning on a light, filling a gas tank, buying groceries or sending a text involves far more interconnected technology than most of us ever see.