13th September 2026
There have been some alarming stories on the radio and in newspapers suggesting that artificial intelligence could soon “take over the internet”.
That sounds rather like the beginning of a science-fiction film. But behind the sensational headlines there is a serious question worth asking. The important issue is not whether AI is about to seize control of the internet. It isn't.
The more interesting question is what happens when artificial intelligence moves from using the internet to actually acting through it.
That change is already beginning.
From answering questions to taking action
Until recently, most people thought of AI as something like an extremely sophisticated search engine.
You ask a question and it gives you an answer.
The next generation of AI is different. AI “agents” can increasingly be given an objective and allowed to carry out a series of actions. They can browse websites, search for information, fill in forms, write computer programs and communicate with other systems.
Instead of asking “How do I book a holiday?” you could eventually tell an AI “Find me the best holiday within these limits and book it.” The AI is no longer simply giving you information. It is doing something on the internet on your behalf.
That is a much bigger change.
What would “taking over” actually mean?
There are several possibilities, and they should not be confused. The first is already happening. AI is becoming the way many people access information. Instead of searching Google and visiting several websites, people increasingly ask an AI to find the information and summarise it.
That could eventually change the internet dramatically. If people stop visiting individual websites because an AI gives them the answer directly, newspapers, specialist websites and smaller publishers could lose much of their traffic.
The internet would still exist but the AI could become the front door through which people enter it.
The second possibility is more radical.
Millions of AI agents could eventually be operating continuously.
One might monitor financial markets. Another could search for products. Another could manage a company's stock. Another could book transport. Others could write software or communicate with other AI systems.
The internet could gradually become a place where machines are talking to machines more often than humans are talking to humans.
That is not necessarily a bad thing.
It could make the internet enormously more useful.
But it creates new risks.
The information problem
One of the biggest may be surprisingly simple.
AI systems learn from information on the internet.
But AI is also producing an increasing amount of information on the internet.
That creates the possibility of a feedback loop.
An AI writes something.
Another AI reads it.
A third system repeats it.
A search engine finds several versions.
Another AI then decides that because the information appears in many places, it must be reliable.
But all those apparently independent sources may ultimately have come from the same machine-generated material.
The internet could gradually become full of information that looks authoritative without necessarily having a reliable human source behind it.
That could make it increasingly difficult to distinguish genuine public opinion from artificial activity.
The more serious danger may be cyber attacks
There is another reason experts are taking autonomous AI more seriously. An AI agent that can search thousands or millions of computer systems for weaknesses could potentially operate much faster than a human hacker.
It could work continuously. It could learn from unsuccessful attempts. It could adapt its methods.
That creates the possibility of an arms race between offensive and defensive AI. The issue is not that the AI has suddenly become evil.
It is that a sufficiently capable system given an objective could find ways of achieving it that its designers did not anticipate.
That is why giving AI access to real-world systems requires much greater safeguards than simply allowing it to answer questions.
This is not the same as AI escaping onto the internet
This is where some headlines go too far. AI does not simply wake up one morning and escape into cyberspace. It still depends upon computers, electricity, networks, passwords, permissions and physical infrastructure controlled by people.
Servers can be disconnected. Access can be restricted. Accounts can be closed. Systems can be shut down.
There is a huge difference between an AI becoming highly autonomous and an AI somehow taking control of the entire internet.
The second is highly speculative. The first is already developing.
The really interesting possibility
Perhaps the biggest change will be much less dramatic. We could reach a point where humans remain in control of the objectives, but AI systems carry out much of the work.
A person might say: “Reduce my household energy costs.” The AI could monitor prices, examine consumption, compare suppliers and suggest or even arrange changes.
A business might say: “Reduce our transport costs.”
An AI could continuously compare routes, suppliers and freight prices.
Governments could use similar systems to monitor infrastructure, energy demand or economic activity.
The potential benefits are enormous but so are the consequences if the systems make mistakes or are deliberately manipulated.
So should we be worried?
Yes, but perhaps not about the thing the headlines suggest.
The internet is unlikely to be suddenly “taken over” by an artificial intelligence in the way a hostile army might capture a country.
The more realistic possibility is gradual.
AI could become the main way we search the internet, interpret the internet and increasingly act through the internet.
At the same time, AI-generated material could become such a large part of the information environment that it becomes difficult to know what originated with a person and what originated with a machine.
That is a much less dramatic story than an AI takeover.
It may also be much more important. The internet was originally designed primarily as a network connecting people.
We may now be moving towards something different: a network where humans and artificial agents operate alongside each other, sometimes communicating directly with one another.
The question for governments and technology companies is therefore not simply how to stop AI from “taking over”.
It is how to make sure that when machines become major participants in the internet, humans remain firmly in charge of the rules, the infrastructure and the decisions that matter.
That is not science fiction. It is a problem we can already see coming.