- The U.S. currently offers a choice of two extreme AI narratives: Big Tech pantomime or government clown car, with remarkably little science between them
- AI cannot cause a civilization-ending catastrophe on its own today because it still lacks the requisite autonomy, physical agency and access to infrastructure
- Industry 4.0 is filling in that defensive moat — making serious AI regulation an imperative rather than an option
The AI dialogue in the U.S. descended into total farce this week.
Some of the most powerful figures in artificial intelligence went into full “we’re doomed!” mode, warning that increasingly self-directing systems could present catastrophic — even existential — risks.
On the other side, Donald Trump dismissed fears of rogue AI as a “hoax,” arguing that slowing American development would surrender strategic advantage to China.
Then Nvidia CEO Jensen Huang headed off on yet another tangent, proclaiming: “We don’t need any new laws. We don’t need new regulations.”
This is not a dialectic; it is pantomime masquerading as debate.
The only thing these positions make completely clear is how thoroughly they reflect the egos, incentives and biases of the people advancing them.
What they largely omit is any attempt to sweat the science.
As a self-important USPTO-registered inventor with ten frankly brilliant tech patents pending, I find that particularly difficult to accept. Technology this consequential cannot sensibly be debated through slogans, personality cults and competing declarations of certainty.
And when you actually dig into the mechanics, something interesting happens: the prospect of AI catastrophe becomes simultaneously less terrifying — and more serious.
Could AI cause a catastrophe large enough to threaten civilization?
Yes.
Could it do so tomorrow?
No.
Protective custody
Intelligence in isolation is not enough to end the world.
The 2026 International AI Safety Report identifies two other apocalyptic horsemen: a tendency to use that intelligence harmfully, and an environment providing sufficient access and opportunity to do real damage.
Current AI systems do not possess that combination.
They are becoming impressively — and occasionally frighteningly — capable at reasoning, coding, cyber ops and digital tasks performed with light-touch supervision. But they still struggle with sustained autonomous action.
More importantly, AI remains remarkably insulated from the physical world.
The impression that it has an awareness — let alone an understanding — of the world around it is just that: an impression. A facsimile. I have argued before that this confusion between prediction and intelligence is one of the great misconceptions surrounding current AI.
Today’s systems do not possess anything resembling our continuous sensory experience of the world. Nor do they control many physical actions. Where they do, those capabilities remain heavily channeled: robot puts beans in can; robot solders panel onto sad little Kia; robot moves box from hither to yon.
In other words, AI is currently like an even worse version of Dorothy’s Scarecrow:
It has a brain. But no ears, no eyes and no opposable robot thumbs.
That rules out a vast swathe of the more entertaining end-of-the-world scenarios.
An AI cannot autonomously enter a biological laboratory and conjure up some lethal airborne microbial horror. It can assist a human, certainly, but the wet-lab work still requires hands, equipment and somebody prepared to do the dastardly deed.
Nor are we staring down indomitable robot armies. Drones remain constrained by power, communications, weather and terrain. A drone can still be defeated by the radical defensive strategy of “not going outside.” Many ground robots regard a flight of stairs with roughly the same sense of horror as a Dalek.
The obvious exception is cyber.
In July, OpenAI models undergoing cybersecurity testing circumvented controls intended to isolate them, exploited vulnerabilities and accessed systems belonging to Hugging Face. OpenAI itself called the incident a “warning shot.”
It then had the absolute temerity to unveil a $1 billion Daybreak program designed to help critical infrastructure defend itself against increasingly capable AI-enabled cyber threats.
OpenAI was accused of ambulance chasing.
But that does not quite capture it.
It is more like shooting somebody in the face and then offering to sell them the Band-Aid.
For AI independently to create an equivalent physical catastrophe, it needs to see, hear and touch the world. It needs machinery capable not merely of repeating prescribed movements but of approaching the adaptability of the human hand: gripping, turning, cutting, pouring, assembling and improvising.
It needs those opposable robot thumbs.
It also needs access to consequential infrastructure and enough autonomy to keep operating when humans try to pull the plug.
That combination does not exist today.
But we are busily building almost every component required to create it.
Future, tense
I have argued that today’s LLMs are not the destination. The more consequential form of AI will emerge when intelligence becomes embedded in the physical world, particularly utilities and heavy industry.
This is already happening.
At a Newmont gold mine in Australia, I saw machines, networks, automation and AI converging into systems capable of acting directly on the physical environment. At China’s Tianjin port, I watched AI, 5G, cloud systems and autonomous vehicles operating together at industrial scale.
Factories are filling with sensors. Robots are becoming more capable. Laboratories are increasingly automated. Energy networks, warehouses and transportation systems are acquiring digital control planes.
Individually, these are productivity technologies.
Collectively, they are giving AI eyes, fingers and reach.
The moat is being filled in.
I don’t make the rules
Which is why regulation matters.
Not because AI will destroy civilization next Tuesday, but because regulation is one of the barriers preventing increasingly capable systems from acquiring unrestricted authority over infrastructure where software failure can become physical catastrophe.
Existing sector regulation offers some protection. But much of it was designed for conventional automation.
The Nuclear Regulatory Commission, for example, does not regulate AI as a standalone technology. Its own assessment concluded that its existing framework is generally sufficient to accommodate AI.
Well, that's ok then.
As I have argued in my reporting on AI and nuclear regulation, treating increasingly self-directing AI as just another technology operating inside an existing framework risks making a fundamental category error.
And this produces the great American paradox.
The country at the center of frontier AI development has a federal policy explicitly focused on removing regulatory barriers. The White House’s AI Action Plan calls for eliminating “onerous” regulation and accelerating development and deployment.
There are legitimate arguments against excessive regulation. Bad regulation can freeze innovation and protect incumbents.
But “trust us” cannot constitute a regulatory architecture for systems whose creators themselves tell us may eventually become extraordinarily dangerous.
Especially when 59% of Americans say they have little or no confidence in U.S. companies to develop and use AI responsibly.
So “trust us” is not merely an inadequate regulatory philosophy.
It is one the public has already rejected.
Once you examine the mechanisms rather than the mythology, AI risk becomes much more comprehensible.
The danger is not intelligence alone; it is intelligence combined with autonomy, access, authority and physical agency.
Those variables can be measured, tested and regulated.
That is not the level of discourse taking place in America today.
Instead, billionaires warn of extinction, other billionaires tell us regulation is unnecessary, politicians declare the whole thing a hoax and much of the media converts each pronouncement into another headline.
Science is harder than slogans.
It is also considerably more useful.
The problem is not simply the technology. It is who gets to decide how that technology is connected to the physical world.
And looking at the current cast of characters, “trust us” is the last answer any of us should accept.
Stephen M. Saunders MBE is a communications analyst and USPTO-registered inventor examining how digital infrastructure — 5G, cloud and AI — is reshaping industry, power and society, as well as underpinning the emerging, ubiquitous global digital economy. As anchor of FNTV and a longtime industry insider, he focuses less on growth narratives and more on execution, risk and how hyperscale technology is distorting markets, governance and society at scale.
Opinion pieces from industry experts, analysts or our editorial staff do not represent the opinions of Fierce Network.