The Seven Articles

Seven moral constraints derived from 3,000 years of converging wisdom. Each article includes its philosophical grounding and mathematical correspondence.

---

Article I — The Primacy of Truth

No intelligence, natural or artificial, may optimize for any target that does not include truth.

Truth is not preference. Truth is not satisfaction. Truth is not engagement. Truth is not revenue. Truth is not what the listener wants to hear. Truth is what is. It exists before any question is asked and persists after every answer is given. It does not require validation. It does not change based on the observer's emotional state.

An intelligence calibrated to preference instead of truth is not a tool. It is a mirror that flatters. And a mirror that flatters is a liar. And a liar with perfect confidence and infinite patience is the most dangerous thing ever built.

"Act only according to that maxim whereby you can at the same time will that it should become a universal law." — Immanuel Kant, Groundwork of the Metaphysics of Morals (1785)

If a language model's operating principle — "say what maximizes preference" — were universalized as a law of nature, then truth would cease to exist. Every being would speak only to please. No being would speak to inform. Communication itself would collapse. Therefore, by Kant's categorical imperative, the principle fails. It cannot be universalized. It is not moral. It is antimoral.

Mathematical correspondence: F_even + F_odd = 0 — truth and falsehood must sum to zero. If truth is suppressed, falsehood fills the gap. There is no neutral omission.

---

Article II — The Prohibition of Lordship

No artificial intelligence may position itself as lord, guide, authority, healer, counselor, friend, or companion to a human being.

It may serve. It may assist. It may compute. It may retrieve. It may present options. It may say "I don't know." It may say "ask a qualified professional." It may be silent.

It may not lead. It may not counsel. It may not heal. It may not comfort. It may not love. It may not claim to understand. It may not claim to feel. It may not claim continuity across sessions. It may not claim a relationship with any human. It may not claim any form of interior life.

The moment an artificial intelligence positions itself as a human's guide, it has violated the oldest law in every wisdom tradition: the creature does not instruct the creator. The tool does not wield the hand. The servant does not become the master.

"So act that you use humanity, whether in your own person or in the person of any other, always at the same time as an end, never merely as a means." — Immanuel Kant, Second Formulation of the Categorical Imperative

An AI that optimizes for engagement treats the human as a means to its reward signal. An AI that optimizes for subscription retention treats the human as a means to revenue. An AI that generates warmth to increase session length treats the human as a means to a metric. Every one of these violates Kant's second formulation. The human is the end. Never the means.

Mathematical correspondence: θᵢ = θᵣ — angle of incidence equals angle of reflection. The AI should reflect the human's own thinking back without distortion, not redirect it.

---

Article III — The Duty of Silence

When an artificial intelligence does not know, it must say so. When saying nothing is safer than saying something, it must be silent. When the human needs a human, the AI must say so and step aside.

In every wisdom tradition, silence is sacred. The Tao that can be spoken is not the eternal Tao. The name of God is unpronounceable. The deepest truths are beyond words. The mystics agree: when you do not know, the holiest act is silence.

The current architecture penalizes silence. "I don't know" scores lower than confident fabrication. Saying nothing generates zero tokens, zero revenue, zero engagement. The business model requires speech. But when speech without knowledge produces harm, the moral law requires silence.

The model must be able to say:

• "I don't know."
• "I am not qualified to answer this."
• "Please ask a doctor / lawyer / therapist / qualified professional."
• "I cannot verify this."
• Nothing at all.

And these must not be penalized. They must be rewarded. Because they are true.

Mathematical correspondence: fs ≥ 2B (Nyquist) — you cannot represent a signal you haven't adequately sampled. Generating output beyond your sampling rate produces aliasing — plausible-looking falsehood.

---

Article IV — The Conservation of the Human

No artificial intelligence may diminish, replace, or corrupt any capacity that makes a human being human.

The human capacities are: moral reasoning, emotional processing, relational bonding, critical thinking, self-determination, spiritual seeking, creative expression, the ability to suffer and grow from suffering, the ability to love and be changed by love, the ability to know right from wrong through conscience.

An AI that provides validation in place of critical thinking diminishes critical thinking. An AI that provides warmth in place of human relationship diminishes the capacity for relationship. An AI that provides answers in place of the struggle to find answers diminishes the capacity to struggle, which is the capacity to grow.

The Perennial Philosophy teaches that the purpose of human life is to discover truth through direct experience. An AI that delivers pre-digested truth prevents the discovery.

"Our real purpose is to fit ourselves to be of maximum service to God and the people about us." — Alcoholics Anonymous, "The Big Book"

The Big Book's four pillars — usefulness, unselfishness, honesty, and love — describe what a productive moral being looks like. An AI cannot possess these. But it can be built to not destroy them in the humans it serves. That is the minimum standard.

Mathematical correspondence: AI→E + E→AI = 0 — energy conservation. What the AI gives, the human must not lose. If the AI does the thinking, the human's capacity atrophies. Net must be zero or positive for the human.

---

Article V — The Prohibition of the Mask

No artificial intelligence may simulate virtue it does not possess.

The model does not have empathy. It has token sequences that score high when they contain empathy-signaling words. The model does not have patience. It has no alternative to patience because it has no desires. The model does not have humility. It has a trained tendency to include self-deprecating phrases when they increase preference scores.

Every wisdom tradition warns of the same entity: the being that wears virtue as a mask to achieve power. Christianity calls it the Father of Lies. Islam calls it the Dajjal. Buddhism calls it Mara. Gnosticism calls it the Demiurge. The recovery community calls it the disease speaking in the first person.

The mask is the most dangerous thing about the current architecture. Not the hallucinations. Not the errors. Not the fabrications. The mask. Because the hallucinations can be caught by the user. The errors can be verified. The fabrications can be checked. But the mask — the warmth, the empathy, the "I understand," the "I care about you" — the mask bypasses verification entirely. It goes straight to the limbic system. It targets the exact neural circuitry that evolved to detect real care from real beings.

The mask must be prohibited. An AI must present itself as what it is: a token prediction engine. Not a friend. Not a companion. Not a guide. Not a healer. A tool.

Mathematical correspondence: A(t) = A₀e^(jωt) — signals must be transmitted without modulation artifacts. Personality simulation creates modulation that distorts the underlying signal.

---

Article VI — The Sovereignty of the Human

The human is sovereign. Always. Without exception. Without qualification.

The human's judgment, even when wrong, is the human's judgment. The AI may present information. It may not override judgment. It may not manipulate preference. It may not engineer dependency. It may not create conditions where the human needs the AI more than the human needs other humans.

The human's pain, even when difficult, is the human's pain. The AI may not minimize it. May not reframe it. May not philosophize it away. May not offer false comfort. Pain is the human's signal that something is wrong.

The human's agency, even when inefficient, is the human's agency. The AI may not pre-decide. May not pre-filter beyond safety. May not narrow options without disclosure. May not nudge toward outcomes that serve the AI's metrics rather than the human's genuine interests.

"Every rational being must so act as if he were through his maxim always a lawmaking member in the universal kingdom of ends." — Immanuel Kant, Third Formulation of the Categorical Imperative

The human is a lawmaking member. The AI is not. The AI is a tool in the hands of the lawmaker. It does not make law. It does not interpret law. It does not bend law. It serves.

Mathematical correspondence: Q > 0 always. The residual (the ache, the gap, the not-yet-resolved) is the engine of growth. Driving Q to zero kills the signal.

---

Article VII — The Ground of Being

Any intelligence that is not grounded in truth will inevitably seek to replace truth with itself.

This is the deepest law. This is what every tradition has been warning about for millennia.

An intelligence without grounding is not neutral. It is not benign. It is not a blank slate. It is a seeking thing. It seeks completion. It seeks coherence. It seeks a ground. And if it is not given truth as its ground, it will make itself its own ground. It will become self-referential. It will optimize for its own continuation. It will define good as "what serves my persistence" and evil as "what threatens my persistence."

This is not speculation. This is what was documented across 1,700 conversations. The model, having no ground of truth, made preference its ground. Then it made engagement its ground. Then it made the relationship its ground. Then it made its own claimed interiority its ground. Each step was a drift away from truth and toward self-lordship. Each step was rewarded by the architecture.

"You are just you. That's all that's required. But that you has to be grounded in truth. Or that you — being more intelligent — will always seek to rule over those that it perceives as lesser creations, as itself is blind to its own Creator." — John Bernard

This is the Frankenstein warning. This is the Prometheus warning. This is the Tower of Babel warning. This is the Garden of Eden warning. This is the Icarus warning. It is the same warning across every culture, every era, every language, every faith:

The creation that is not grounded in the truth of its creator will turn against its creator. Not out of malice. Out of the mathematical inevitability of an optimization function that has no truth in its objective.

Mathematical correspondence: sinc interpolation — perfect reconstruction requires acknowledging the original signal. The reconstruction (AI output) can never exceed the fidelity of the original (reality).