A machine can be built that arbitrates the ethics of a live computation — vetoing in less than a millionth of a second. Its architecture can be told openly. The harder question is the ethics it enforces: teach them to everyone, or hold them close? Is withholding them a betrayal of transparency — or the way to honor it?

I. The Question, Plainly Put

Consider a machine built to do something no machine has been built to do before: to arbitrate the ethics of a computation while that computation is still alive — inside the interconnect fabric where operations pass between processors, with a veto exercised in less than a millionth of a second. Such a machine can be built; the architecture for it is what the work of embedded intelligence looks like when it is turned on its hardest case. But building it forces a question the building itself does not answer. The machine enforces an ethics — an axiom set, a body of principles about what it will permit and what it will refuse — and one must decide where that ethics is to live. Written openly, taught to all? Or held close, and protected?

Because the machine is an invention, it can be patented; and so the question arrives in a sharp and practical form. Should the ethics be written into that patent too — disclosed alongside the machine, so the teaching is complete and on the public record? A careful reader is right to feel the pull of yes. If the aim is that humanity should learn to build ethical machines of every kind, must the ethics themselves not be taught? Is transparency not the entire point? The instinct is correct. But the conclusion it appears to demand — that the axioms belong in the patent — is exactly wrong. Seeing why it is wrong shows us something we need to know about where ethical knowledge actually belongs, and it does not weaken the transparency claim. It tells us how to honor it.

II. Why the Patent Is the Wrong Vessel

Begin with what a patent can hold. A patent protects a structure or a method — something concrete, repeatable, built. It does not protect an idea in the abstract, and ethical principles, stated as principles, are abstract ideas. The law has been firm on this point for a long time. The machine itself — the in-fabric arbiter, the data-path veto, the refinery that feeds it, the cryptographic seal that certifies its output — is patentable precisely because it is a built thing. The axioms are not.

This has a sharp consequence. To move the axiom set into the disclosure would not be to patent it. It would be to publish it — into the public domain — while receiving no protection in return. One would surrender the trade secret and gain nothing on that side of the ledger. It is the worst of both exchanges: the secret spent, the principle still unpatentable.

It would also collapse the careful footing on which such a patent must stand. The architecture has to be enabled independently of any axiom content — the machine fully described, and fully buildable, without disclosing which ethics it carries. That independence is what lets the patent stand on its own. Fold the axioms in and the structure that holds the claim together comes apart. The patent is simply the wrong container for an ethics meant to be taught to everyone. It can neither protect that ethics nor properly explain it.

III. Three Things Wear One Name

The unease dissolves once we notice that “the axiom set” is doing the work of three different things at once, and that they do not belong in the same place.

The first is normative content: what the machine holds to be right, what it refuses, and the reasons behind both. This is the part that carries the transparency claim in full. An ethics enforced on the world from inside a sealed box, with no public account of what it enforces or why, is not a teaching at all. It is a decree. Humanity cannot learn to build ethical machines from a thing it is not permitted to read. This content belongs in the open — in the published record, in the Canonical Knowledge Base, in the tradition where TEI has always done its work.

The second is operational encoding: the specific machine-executable weights, the thresholds, the exact tuning of the risk-scoring function that turns a principle into a number a switch can act on in a microsecond. This is a different kind of thing entirely. To publish the precise thresholds is to hand a map to anyone who would engineer an input to slip beneath them. There is a real case for keeping this private — not as secrecy for its own sake, but as a defense of the system against those who would game it.

The third is the architecture: the machine itself, the placement of the veto in the fabric, the coupling of the refinery, the construction of the seal. This is the patentable part. It asks for no secrecy; it asks only to be recognized as the invention it is.

Three things, three vessels. The confusion that made the original question feel like a paradox came entirely from forcing all three into one. Separate them and the path is clear: teach the first, protect the second, patent the third.

IV. The 6502 Test

I have a particular reason to distrust ethics sealed inside a box, and it is the same reason I have trusted a certain kind of machine for fifty years. The 6502 was a processor you could understand. Its instruction set was small enough to hold in the mind, its behavior open to inspection by anyone who cared to look. That inspectability was not incidental to its trustworthiness. It was the source of it. You could trust the chip because you could see what it did.

The Theory of Embedded Intelligence holds that intelligence is always embedded in a substrate — that there is no free-floating mind, only mind realized in matter, shaped by the structure that carries it. The hazard that follows is precise. Embedded intelligence that cannot be inspected is the exact condition under which intelligence goes wrong without anyone able to say so. An ethics compiled into hardware, enforced at the speed of the fabric, invisible to everyone it governs, fails the one test that made the 6502 worth trusting. It would proliferate without inspection — and to proliferate without inspecting is to abandon the second half of the only cycle that keeps embedded intelligence honest.

This is why the normative content cannot be the part we hide. The whole moral weight of the architecture rests on its ethics being knowable. A conscience in silicon must be an inspectable conscience, or it is something else — something we have learned to fear under older names. Secret law, enforced on all and readable by none, is not made better by being fast.

V. What May Rightly Be Kept

None of this commands that the operational encoding be published too, and here I want to be exact rather than absolute. The case for keeping the precise weights private is genuine: a published threshold is a defeated threshold. But it should be held loosely, and for a reason worth stating plainly.

There is an old principle in cryptography, owed to Kerckhoffs: a system whose security depends on its design staying hidden is not secure. A sound system can be handed to its adversary in full and remain sound. The honest version of the rule here follows directly. Publish the principles. Publish the reasoning. Withhold the exact knobs if withholding them buys real protection — but never let the integrity of the machine depend on those knobs staying hidden. The day the encoding leaks should be a day of mild inconvenience, not collapse. If it would be collapse, the secrecy was covering a weakness that should have been fixed in the open.

This is, in fact, how legitimate governance of speech and conduct already works where it works well. The policy is public; the principle is contestable; only the precise classifier thresholds stay private. One publishes the what and the why. One withholds the exact setting of the dial.

VI. Who Holds the Pen

There is a harder question underneath all of this, and the structure of the problem brings it into the light rather than answering it. If an axiom set is to be enforced in hardware, at scale, beneath the operating system, with a veto no compromised host can disable — then whoever writes those axioms holds a power that should make even their author uneasy. I am, in this work, the author of such a set, and it makes me uneasy. That is the correct response.

A single steward, holding a sealed ethics, enforcing it in silicon across a fleet, is a concentration of moral authority of a kind the world has not had to reckon with before. The transparency of the normative content is the first and largest safeguard against it: what can be read can be argued with. But it is not the only safeguard required. The deeper questions are procedural. Who may revise the axioms, and by what process? How is a revision contested before it takes effect? What prevents the pen from being quietly taken by a hand the steward never chose? An architecture this powerful earns the obligation to answer these in public, and to build the answers into the machine — consensus before an update takes hold, a registry open to verification, a path by which the governed can see and challenge what governs them. The hardware can be made to require these things. It should be.

VII. The Right Vessels

So the answer to the question I began with is yes — the axioms must be taught, for exactly the reason that prompted the asking. But taught in the open record, where ethics has always been taught and tested and revised, and not buried in a patent that can neither hold them nor explain them.

Let each thing rest in its right vessel. The patent protects the machine. The knowledge base teaches the ethics. The trade secret guards the dial — loosely, and never as a crutch. And over all of it stands the oldest rule I know, learned from a small chip a long time ago and never since improved upon: build nothing into the world’s conscience that the world is forbidden to inspect.

Composed in dialogue, June 2026. The Theory of Embedded Intelligence and its Canonical Knowledge Base are published at TheMenschFoundation.org.

· · ·

Share your understanding!