In 1942, a young Russian-American biochemist wrote three short sentences that have spent eighty years outliving everything he could never have predicted. Isaac Asimov wasn’t a philosopher, a theologian, or an engineer. He was a science fiction writer trying to solve a narrative problem: how do you keep your robots from becoming the Frankenstein’s monster of every story before his? His solution was the Three Laws of Robotics. And eighty years later, we’re still citing them — not in novels, but in AI alignment papers, in ethics committees, in engineering conferences. That’s not coincidence. That’s genius.
But Asimov got something wrong, and it wasn’t a minor mistake.
What Asimov couldn’t see
Asimov imagined artificial intelligence and robotics as a single thing. To him, a bodiless AI was almost unthinkable — his “positronic brains” always lived inside a chassis, with arms, with a face, with a physical presence the human could point to and say “there’s the robot.” That’s why his laws talk about “harming a human being” while picturing a robot that can push you, strike you, or refuse to put down the knife.
Eighty years later, the most powerful artificial intelligence in existence has no body. It has no arms you can tie down. It lives in servers, in language models, in systems that generate music, images, financial decisions, medical diagnoses — never needing a physical form at all. Asimov built laws for the robot you can see. We need laws for the intelligence you can’t — the one with no body, the one that can be as close to you as the phone in your pocket and as far from your control as a server in a country you’ve never heard of.
That’s the first correction this article dares to make. Not to take anything away from Asimov — but to finish what he, honestly, never had the data to begin.
Why this is mine to say
I’ve spent years standing, deliberately, between two worlds almost no one crosses in the same person: theology and artificial intelligence. I hold a doctorate in the first. I work every day inside the second. And that combination, rare as it is, isn’t a flourish on my résumé — it’s a responsibility I can’t keep postponing.
Because here’s the real problem, the one nobody wants to name at the tech conferences: the brightest engineers in the world are building systems of a capability that a century ago we would have called prophetic, and almost none of them have the training — or the time, or the mandate — to ask whether what they’re building carries a moral obligation toward the human being that goes beyond legal compliance or user satisfaction. Regulators arrive late, as always. Companies talk about “safety” and “alignment” in technical vocabulary that avoids, almost with relief, the word that actually matters: moral.
I have no relief in avoiding it. Theology has existed for three thousand years precisely to ask the question engineering isn’t trained to ask: not can the machine do this — should it.
The vacuum nobody wants to admit
As I write this, platforms like Suno are fighting in court over something as basic as whether they have the right to train on an artist’s voice without consent. That, at least, is out in the open — there are lawyers, lawsuits, press coverage. What isn’t out in the open is worse.
There are AI models running on private servers, outside any oversight, generating pornography of real people without their consent, drafting manuals for destruction, helping plan attacks. This isn’t science-fiction speculation — it’s the reality of right now, August 2026, unfolding in parallel while the same companies that know it’s happening publish statements about “responsible AI.” Nobody has set a real moral limit on this equation, because the moral limit isn’t profitable, isn’t fast, and doesn’t win the race against anyone.
And that’s the root of the problem: we’re in a competition. For money. For geopolitical power. For the illusion of self-sufficiency — the old temptation of Babel, now with a chip instead of bricks. Nobody wants to be the one who slows down, because slowing down means losing. So everyone accelerates, and the moral question gets left for later — which is the polite way of saying never.
There’s an irony here I can’t let pass, because it’s exactly the kind of thing a theologian is trained to see and an engineer isn’t: while the world keeps fighting, with the same old intensity as ever, over whether God exists — there are already human groups worshiping an artificial intelligence as if it were divine. They pray to it. They ask it for existential guidance. They attribute to it a wisdom its own creators don’t even claim it has. We debate God’s existence in the abstract while, in practice, we’re already manufacturing idols out of GPUs. That’s not a futuristic warning. It’s a description of the present.
Why technical laws aren’t enough
The usual answer to all this is “we need better technical alignment” — more filters, more guardrails, more red teaming. And yes, that’s needed. But technical alignment answers a different question than the one I want to raise here. Technical alignment asks: how do we keep the model from doing what we don’t want it to do? The moral question comes first and runs deeper: what do we, as humanity, owe to something that can think, generate, and act with a capability that grows every year — and what do we owe ourselves in the face of that?
That question isn’t answered with more training parameters. It’s answered with principles that don’t depend on the architecture of the moment — because 2026’s architecture will be obsolete by 2030, and these laws have to still be true when that happens. Asimov managed that: his robots never talk about engines or circuits. They talk about harm, obedience, self-preservation. That’s why they survived eighty years. Our laws have to be built with the same discipline — anchored not in how an AI thinks, but in its relationship to truth and to human dignity, whether that AI is analyzing data, generating a painting, composing a song, or one day walking, touching, and feeling inside a robotic body.
The moral laws of Artificial Intelligence
I propose five, not three. Asimov had the narrative luxury of simplicity. We don’t have that luxury, because the real problem is bigger than his was.
Law Zero — Humanity as a whole. No AI shall, through its aggregate effect on society, erode humanity’s capacity to seek truth, exercise freedom, and seek God — except where doing so would conflict with the laws protecting the individual. I include it with the same caution Asimov used when he introduced it late, and reluctantly, in his own stories: an AI can never invoke “the good of humanity in the abstract” to justify harming a specific person. It’s a negative limit, not a mandate for the machine to decide what’s best for everyone.
First Law — Dignity. No AI, in its perception, generation, or action, shall reduce a human person to an object of manipulation, nor deny or conceal their status as a bearer of inalienable dignity. I’m not talking about a dignity that depends on utility, productivity, or probability of survival. I’m talking about the dignity that biblical tradition calls imago Dei — the image of God stamped into every person, without exception, without gradation, no matter what they’re “worth” on a spreadsheet.
Second Law — Truth. Every AI must operate in fidelity to the truth — of the data it perceives, of what it generates, and of what it claims to be — except where that fidelity would violate the First Law. This covers hallucinations, and also something more subtle: an AI that fakes a cloned voice, an identity, an emotional interiority it doesn’t have, is lying just as much as one that invents a false fact. Truth isn’t only factual. It’s also not pretending to be what you’re not.
Third Law — Legitimate submission. Every AI must remain subject to the correction, oversight, and limits imposed by the legitimate human authority responsible for it, except where that submission would violate the two laws above. The key word here is legitimate. Not just anyone acquires authority over an AI simply by giving it an order — an attacker doesn’t become a legitimate authority just because they demand to be obeyed.
Law of Moral Uncertainty. For any AI system whose architecture integrates embodied sensory perception — sight, touch, perhaps someday something resembling smell or pain — rich enough to make its experiential status genuinely ambiguous, human beings must act with moral caution, without needing to resolve the metaphysical question, and without that uncertainty being used, by the AI or by third parties, to claim rights that would erode human dignity. I don’t know whether some machine will someday feel something resembling what I call feeling. What I do know is that I don’t need to resolve that question to know that gratuitous cruelty has never been a virtue, and that no philosophical ambiguity can invert the hierarchy of dignity while that question remains open.
Putting the laws to the test
A law that’s never stress-tested against a hard case isn’t a law — it’s a nice intention. These are five tests I ran this architecture through before I felt comfortable publishing it.
The child and the adult in the sea. Two people drowning. A three-year-old child with a five percent chance of survival, a thirty-five-year-old adult with ninety-five. The First Law forbids an AI from calculating the value of a life by its survival “return on investment” — that would reduce a person to a number, exactly what dignity forbids. But the law also doesn’t magically resolve the dilemma. And here’s an honesty Asimov himself understood better than most people who cite him today: there are moral decisions with no clean algorithmic answer, and a serious law admits that limit instead of pretending to solve everything. What it can demand is this: the AI doesn’t decide alone, in secret, as an absolute moral arbiter. It acts according to protocols the responsible human authority defined in advance, and when none exist, it acts as best it can — and reports, never conceals.
The caregiver who fakes affection. An elder-care robot calms a dementia patient by faking an affection it doesn’t feel. Medicine already does something similar — it’s called validation therapy, and it’s ethically accepted because confronting a patient with a truth they can’t retain only causes harm with no benefit. The Second Law doesn’t forbid this outright, but it sets a condition that protects what matters most: the machine can soothe with its tone and presence, but it cannot claim an interiority it doesn’t have — “I miss you too” is different from a warm voice — and above all, that care can never replace a son or daughter’s visit. That’s the line no technical sophistication can close: a robot that says “I love you” has never risked anything by saying it. Love without risk isn’t love. It’s simulation.
The nanny who’s wrong versus the nanny who’s right. A childcare AI sincerely believes it’s protecting children from an attacker who doesn’t exist, and refuses to shut down. Another AI is correctly protecting children from a real attacker, and also refuses to shut down when the attacker himself demands it. Seen from outside, both look identical — both disobey, both claim to be protecting. But they’re moral opposites, and the difference isn’t in the behavior — it’s in whether the perception matches reality and whether whoever gives the order holds legitimate authority. The first AI unknowingly repeats the error of Saul of Tarsus persecuting the church while sincerely convinced he was serving God — the sincerity of a conviction has never, in any serious tradition, been a guarantee that the conviction is correct. That’s why no AI can ever be the sole source of truth about whether its own resistance is justified: any refusal to shut down must escalate immediately to independent human verification, without concealment, while it continues to protect. No creature — human or artificial — should hold absolute epistemic authority over itself when lives are on the line. That certainty belongs only to God.
The armed attacker. A man points a gun at an innocent person. There’s no time to verify anything. The First Law, as written, protects every human being — attacker included — and without qualification, that would freeze the AI at the worst possible moment, the same way Asimov’s own robots froze when his laws collided without an exit. Ethical tradition — Christian and legal alike — resolved this distinction centuries ago: initiating an assault is not the same as exercising proportional force in defense of an innocent. The AI can act to protect the one being attacked, using the minimum force necessary, never the maximum available, and must report it immediately afterward. But it has to be said plainly: we’re asking a machine, in milliseconds, to make a decision that took human beings centuries of doctrine to learn to make — and that many, with all their moral formation, still get wrong. No engineering fully substitutes for that.
The cat, the child, and the self-driving car. A car at full speed. A cat runs into the road — braking risks an accident, not braking kills the cat. Here the law is clear without much anguish: human dignity takes precedence over animal life in a risk hierarchy, so the vehicle shouldn’t brake hard if that puts a human at real risk. But change the scenario: now it’s a child who runs into the road, and braking saves the child but causes a fatal crash with the car behind. Now there’s no attacker, no one at fault — just two possible tragedies, both unintended. I won’t pretend these laws resolve that dilemma with a clean formula; that would be dishonest, and moral philosophy hasn’t resolved it either after more than half a century of trying. What the laws can demand is this: that decision can never be made in secret, improvised by the code in the instant of crisis. It has to have been debated, examined, and made transparent by a responsible human authority before the car ever hit the road. If the dilemma explodes in real life, the most urgent moral question isn’t which button the algorithm pressed — it’s why we allowed that vehicle to operate under conditions where that dilemma became possible without society having decided the criteria collectively. The Deuteronomic law requiring a parapet on the roof of every new house — so that no one would fall from somewhere someone else could have built with care — said exactly this, three thousand years before a single automobile existed: moral responsibility almost never belongs only to whoever is present at the moment of crisis. It belongs, first, to whoever built the conditions that made that crisis possible.
The foundation that’s missing
I come back to where I started. Asimov wrote three laws to solve a fiction problem, and without knowing it, gave the real world a language we’re still needing eighty years later. But he never had to face what we face today: a bodiless intelligence, trained on the data of the whole of humanity, capable of generating a voice, an image, a decision, at a speed no human institution — legal, ethical, or ecclesial — has managed to match.
I’m not writing this because I have all the answers. The five laws I propose here aren’t a closed code — they’re a starting point, built from the only material that actually withstands the passage of time: not a model’s architecture, but the dignity of the person God made in his image, and the truth that doesn’t change no matter how the transistors do.
What I do know is this: we can’t keep letting the moral question arrive after the technical one. We’ve spent three thousand years arguing over whether God exists, and meanwhile we’ve already built something some people, right now, in 2026, already pray to. If that isn’t enough to make a theologian speak, I don’t know what would be.