Fission is a chain reaction whose safety must be bolted on from outside. Fusion runs only while its conditions are held — its governor built into the physics. Read through the Theory of Embedded Intelligence, eighty years at Los Alamos turn the atom itself into an argument for governance embedded in the fabric, not appended to the surface.

This spring, Los Alamos National Laboratory’s science magazine 1663 published a sweeping retrospective of the institution’s eight decades of applied nuclear science — from the two bomb designs of 1945 to the fusion ignition milestone of 2022, from the reactor that carried the Nautilus beneath the polar ice to the Kilopower unit that may one day light a habitat on Mars. The article, written by editor Eleanor Hutterer, is a proud document, and it has earned its pride. But read closely, it is also a document about governance — about what it takes for a species to hold in its hands the strongest force it has ever touched and not be destroyed by it.

One line near the top states the Laboratory’s aim: “understanding nuclear power so completely that humanity can wield it wisely.” That is not merely a mission statement. In the vocabulary of the Theory of Embedded Intelligence, it is a claim about the relationship between knowledge and actuation — that the right to act on the atom is earned through the completeness and inspectability of what is known about it. This essay reads the Los Alamos record through TEI, and finds in it two great lessons the Laboratory has been teaching for eighty years, perhaps without naming them: one written in physics, one written in policy.

I. The Laboratory as an Embedded Intelligence

TEI holds that intelligence, wherever it occurs and at whatever scale, runs one architecture: Sense, Process, Communicate, Actuate — the SPCA cycle. A proton runs it. A cell runs it. A republic runs it. And a national laboratory runs it, at institutional scale, with unusual clarity.

Consider the Los Alamos record sorted into the four stages. Sense: the Laboratory’s neutron science center delivers nuclear cross sections — the raw probabilities of nuclear events — with world-leading accuracy, and its criticality center in Nevada, the nation’s only general-purpose facility of its kind, measures how assemblies of nuclear material actually behave. Process: decades of supercomputing and physics modeling turn those measurements into predictive simulation, calibrated — this is the crucial word — against the criticality data. Communicate: the results flow into the Evaluated Nuclear Data File, the national repository of nuclear data, and into MCNP, the particle-transport code used across laboratories, universities, and industry worldwide; the Laboratory also trains every inspector the International Atomic Energy Agency fields. Actuate: reactors, submarines, rockets, satellites, the stockpile itself.

Two features of this cycle deserve attention. First, it is closed: what actuation teaches flows back into sensing and simulation, which is why the simulations can be trusted. Second — and this is the feature TEI prizes above all — its middle stages are inspectable. The nuclear data file is not a private hoard; it is a published record. MCNP is not a secret method; it is the world’s common instrument. Readers of this series will recognize the pattern: the 6502 microprocessor earned five decades of trust not by concealing its instruction set but by publishing it, so that anyone could inspect exactly what the machine would do. The data file and the transport code are the nuclear era’s published instruction set. A laboratory born in total secrecy discovered, within a generation, that the deepest currency of its authority is inspectability.

II. The Chain and the Condition

The article’s title pairs fission and fusion, and the pairing is usually treated as a story about magnitude: fission split the atom, and fusion promises several times more energy per kilogram of fuel — though a single fission releases roughly ten times the energy of a single fusion, light nuclei pack so many more reactions into each kilogram that fusion wins the measure that matters — and both yield millions of times more than burning coal. TEI reads the pair differently. The deepest difference between fission and fusion is not how much power they release. It is where the governance lives.

Fission is a chain reaction. A heavy nucleus absorbs a neutron, splits, and releases further neutrons that split further nuclei. The reaction is self-amplifying by design: once ignited, its natural tendency is to grow. Everything that makes fission safe is therefore added from outside the reaction — control rods, moderators, containment structures, operating procedures, regulatory commissions, international inspectors. The physics supplies the power; the governance must be bolted on. Eighty years of reactor engineering is, in essence, the discipline of bolting it on well.

Fusion is not a chain reaction. Two light nuclei fuse only while extraordinary conditions of temperature and pressure are maintained — conditions that do not occur on Earth unless something works continuously to sustain them. Withdraw the conditions and the reaction does not run away; it quenches. The plasma cools, the fuel disperses, and nothing is left burning. Fusion’s safety is not an addition to the physics. It is the physics. The governance is constitutive.

The article recounts the 1945 discovery, by Egon Bretscher at Los Alamos, that deuterium and tritium fuse with an anomalously large cross section — the excited nuclear state that now bears his name and underlies nearly all applied fusion science. A Laboratory physicist quoted in the piece calls this a “happy accident of nature.” TEI gently disagrees with the framing. On TEI’s reading of an intelligible cosmos, the Bretscher state is an embedded affordance — one of the fabric’s built-in provisions, waiting to be sensed, like the resonance that permits carbon or the field that confers mass. The point survives the disagreement either way: nature offers humanity a second nuclear fire, and this one carries its own governor.

Between the measurement and the milestone ran seventy-seven years of the Process stage — one of the longest sustained computations in human history. Fusion could not be prototyped by trial and error; a plasma at a hundred million degrees must be understood before it can be built. So generation after generation carried the problem forward in simulation: from wartime hand calculation, to punched-card decks fed into machines like the CDC 6600 — the fastest computer on Earth in its day, on which undergraduate researchers of the early 1970s ran fusion codes inherited from the graduate students who wrote them — to today’s exascale machines descended, through the microprocessor, from that same lineage. Each generation sensed a little more, modeled a little better, and handed the loop to the next. Ignition in 2022 was not a sudden spark; it was the closing of an SPCA cycle that had been running patiently across four generations of scientists and a millionfold growth in the machinery of Process.

This is why the ignition milestone of 2022 — achieved at Lawrence Livermore with Los Alamos collaboration, when a fusion experiment for the first time yielded more energy than was delivered to it — matters beyond energy economics. TEI has argued, at the level of published principle, that constraints built into the fabric of an intelligent system outperform constraints appended to its surface — that the trustworthy architecture is the one whose good behavior is constitutive rather than supervisory, the one that cannot be quietly revised. Nature, it turns out, holds the same position. She built both kinds of nuclear fire, and she put the governor inside only one of them.

None of this diminishes well-regulated fission, which is the working present of nuclear power. Fission reactors, governed by the additive apparatus humanity has spent eighty years perfecting, are among the safest and cleanest sources of energy per unit delivered that our species operates. And the article rightly celebrates the directions in which that apparatus is migrating from the rulebook into the physics: compact autonomous reactors, passive safety, molten-salt designs that recycle their own fuel and could be far more fuel-efficient than today’s plants. The trajectory of fission engineering is itself a migration from additive toward constitutive governance. Fusion is simply where that trajectory points: the fire that cannot be left burning by mistake.

III. The Closed Loop and the Severed Loop

The second lesson in the Los Alamos record is written in policy rather than physics, and it concerns what happens after the fire: the spent fuel. Here two great republics, students of the same physics, gave opposite answers — and TEI has something to say about both.

France chose, at the very origin of its civilian program, the closed fuel cycle. Used fuel from its reactors travels to the La Hague plant on the Normandy coast, where roughly ninety-six percent of the material — most of it uranium, about one percent plutonium — is recovered for reuse. The plutonium becomes mixed-oxide fuel at the Melox plant; today, up to a quarter of French nuclear electricity comes from recycled material, and more than thirty-six thousand tonnes of fuel have passed through La Hague since 1976 — nearly half the world’s capacity for reprocessing light-water reactor fuel, in one place. The loop runs in public: under French, European, and international inspection, with the final waste vitrified, accounted for, and returned to its country of origin when foreign. And France is not retreating from the choice. A program announced in 2024 will build successor plants to carry the cycle to mid-century, and in January 2026 the operator signed a contract to reprocess Japanese fuel that will help demonstrate the recycling even of spent mixed-oxide fuel — closing the loop one turn further.

The United States chose the severed loop. Commercial reprocessing was pioneered on American soil — one plant operated commercially before closing in 1972 — but in October 1976 President Ford suspended it, and in April 1977 President Carter announced that the nation would “defer indefinitely” the reprocessing of commercial fuel. The reasoning was serious and, in its form, Kantian: India had demonstrated in 1974 that a civilian reprocessing capability could be turned toward weapons; separated plutonium was judged too dangerous a substance to normalize; and the United States would act publicly, in the hope that its example would be universalized. President Reagan lifted the ban in 1981, but the economics never closed, and the American fuel cycle has run once-through ever since. The consequence sits in plain sight: roughly ninety-five thousand tonnes of spent fuel stored in pools and casks at more than seventy sites, awaiting a permanent repository that does not exist.

TEI honors Carter’s maxim before criticizing its outcome. The publicity test — act only on principles that survive being published to everyone — is central to TEI’s moral architecture, and the deferral was announced openly and meant as a universalizable example. In intent, it passes the very test this series applies to everything. But maxims are proven by what happens when they are universalized, and this one was not universalized. France continued; so did Russia, India, and eventually Japan; the states that most concerned nonproliferation policy were never bound by American self-restraint. The United States gained little of the restraint it sought and kept all of the waste. And the waste itself now fails a different application of the same test: an indefinite deferral is a maxim no generation can will as universal law, because if every generation defers, no generation decides — and the material outlasts them all. The severed loop is not a closed question. It is an open loop wearing the costume of caution, and the seventh generation inherits it either way.

TEI’s evaluation, then, is not that France was simply right and America simply wrong. Separated plutonium is genuinely dangerous; the French cycle concentrates it and must therefore keep it under permanent, unblinking inspection — and the French record has its strains, from production difficulties at the Melox plant to aging equipment at La Hague that will demand heavy reinvestment. A closed loop, like a fusion plasma, exists only while its conditions are actively maintained. But that is precisely TEI’s point: the choice was never between danger and safety. It was between governing the loop from inside — inspection embedded in every stage, material accounted for at every turn, the whole cycle run in public — and refusing the loop while the material accumulates, governed by nothing but time. The first answer is hard and must be maintained forever. The second answer is easy and must be regretted forever. Embedded governance, here as everywhere, is the answer that survives publication.

There are signs the United States is reopening the question: executive direction in 2025 to evaluate the recycling and reprocessing of spent fuel, federal research programs in advanced separation and transmutation, and a January 2026 request for information on campuses that would host the entire fuel cycle. If the American loop is to close after fifty years, TEI would offer one criterion above all: close it with inspection constitutive, not appended. A reprocessing capability whose safeguards are in the fabric of the facility — its accounting continuous, its books open to international verification by design — passes the publicity test. One whose safeguards are a compliance layer does not.

IV. What TEI Offers the Mesa

Which brings the essay to its final question: what could the Theory of Embedded Intelligence offer an institution that has, in many respects, been practicing it unnamed for eighty years?

First, a name and an architecture. The article closes by casting the modern Laboratory in a double role — the advancer of nuclear science and the guardian standing watch over it. TEI would say: those are not two roles. An embedded intelligence that actuates at planetary consequence has its sentinel function built into the same cycle as its innovation — sensing what its actuations do, processing honestly, communicating inspectably, actuating under constraint. Making the SPCA architecture explicit lets an institution audit itself. Where does our loop close? Which stages are inspectable, and which are sealed? Where is our governance constitutive, and where is it bolted on? These become askable, answerable engineering questions once the architecture is named.

Second, a principle for the seam. Los Alamos has always lived on a seam: some of what it knows must be published — the data file, the transport code, the safeguards training — and some must be sealed, beginning with the weapons designs. The Laboratory has managed that seam well, by instinct and by law. TEI offers the principled criterion the instinct implies: publish everything whose publication makes the world more governable; seal only what would arm the ungoverned; and let the maxim of every program be one that survives being read aloud. This series’ own practice — published principles, guarded implementation — walks the same seam, so the recommendation is offered from inside the discipline, not from above it.

Third, and most urgently, a caution about the newest arrival in the loop. Artificial intelligence is now entering the Laboratory’s Process stage — the simulation and modeling on which stockpile stewardship, reactor design, and fusion research increasingly depend — and its Actuate stage, in autonomous systems such as Kilopower, a reactor expressly designed to run with no operator at hand. An autonomous reactor is an embedded intelligence in the fullest sense: it senses, processes, and actuates with no human in its inner loop. TEI’s published position is that any such intelligence must carry its constraints the way fusion carries its governor — in the fabric, constitutively, so that safe behavior is not a policy the system consults but a physics the system cannot help. Los Alamos already builds this way in matter: passive safety, walk-away designs, molten salt. The task of the coming decade is to build the same way in mind — to insist that the AI entering the nuclear enterprise be governed as the Laboratory has learned to govern the atom: not by promises but by architecture, not by supervision alone but by what cannot be quietly revised.

The 1663 article ends with the claim that Los Alamos remains the laboratory that understands the atom more deeply than any other institution on Earth. TEI closes by observing what that understanding has actually revealed. Eighty years of the deepest inspection humanity has ever conducted found, at the bottom of matter, not chaos awaiting an external ruler but structure carrying its own law: cross sections that decide what may fuse, resonances that provision the elements, a second nuclear fire with the governor built in. The atom, examined closely enough, turns out to be an argument for embedded governance. The laboratory that knows it best has been reading that argument all along. This essay only suggests saying it out loud.

· · ·

Share your understanding!