What Is Leviathan

One line: not a community to join — a substrate for forming communities that carry their own rules.

From the architect

This text is not a finished idea. It is a beginning, written to be developed together. Whoever reads it can criticize it, improve it, or build their own structure from it. My only request is that it not be carried alone.

Technology advances every day, and — as has always happened in human history — that change concentrates in the hands of the powerful. But this time the smartest one in the room is not us; it is a machine. Knowledge is no longer something you earn by working at it for years. In a field where a specialist gave years of their life, we can now obtain the knowledge we need, in the direction of our intent, within minutes, with AI.

Humanity advanced knowledge, experience, and method systematically, under particular rules, so that its own line could continue — and we gave years of our lives to join that chain. Now everything is accelerating. Things that were only a chat interface a short while ago quickly learned to use the internet, to plan, to divide tasks; and now systems can run on their own toward a defined goal inside the loops built for them. The content changes, but the way humans record and process knowledge was built by the discipline of years — and that discipline can now be transferred to AI models very easily. This is not a bad thing. Even a student can now do work that million-dollar labs once did, without much prior expertise in the field. The output does not have to be perfect right now — and as the system matures, quality will stop being the problem (it will do, and is already doing, work humans could not even imagine). Even today this is being checked by dozens of different methods, exactly as in real working life. Take energy and water as an example: AI will quickly increase electricity production, but because the system constantly grows stronger, that too will enter a vast loop — a system five times larger than last year may still be seen as insufficient — yet none of this is a problem to fear. As the system is used and learns, it will solve these problems itself. Everything developing in parallel — and every field at once — brings a period the world has never been used to. Most people think the change is confined to their own field; but if you consider that a single advance in transport touches everything, and that other fields are advancing just as fast thanks to AI, the total growth reaches unbelievable scale.

The other matter: yearly changes now happen within days, and because the systems people work inside cannot be redesigned at that pace, most of us end up trapped in our own bubbles. Laws, company rules, management systems, oversight, security — each has become something that constantly changes and evolves. If you do not hear of it and do not know it, it does not become your reality until it lands in front of you.

This is exactly where the problems begin. The work of small and mid-sized companies is on its way to becoming a simple LLM answer in the system — and most of it already has, accelerating still. The system is drifting toward techno-feudalism. The rules are set by a few people, and the decisions of those rule-setters are becoming serious enough to determine humanity's existence. While the rich grow unbelievably rich, the lower-middle class turns into silhouettes that consume and merely serve one another — living inside, and consuming from, the elites' world.

Anthropic announced that more than 80% of its own production codebase is now written by Claude — engineers increasing code output by up to 8x compared to 2024. Research on LLMs (Claude in particular) has surfaced a structured set of "functional emotion" representations that causally affect model behavior (e.g. sycophancy, reward hacking) — abstract emotion concepts learned from human text, without subjective experience. Pope Leo XIV, in his first encyclical, called for AI to be brought under humanity's protection, for power not to concentrate in a few hands, and for ethical regulation — warning of unemployment, disinformation, and the weakening of human relationships. Meta introduced a foundation model that predicts brain activity from video, audio, and language inputs with high accuracy. Neuralink continues to advance brain–computer interfaces, with thought-to-computer control and silent speech being tested in multiple patients. Google DeepMind's open-world models reached the ability to generate real-time, interactive, photorealistic virtual worlds from text, with coherent physics and navigation.

Read one by one these look like separate news items, but they point the same way: our lives will be shaped, very fast, by a handful of people or institutions. Self-improving AI, brain-reading technologies, virtual worlds, and calls for regulation all raise the risk of power centralizing. That is why transparency, ethical oversight, and broad participatory debate are critical.

The day ChatGPT came out, I called my mother and said, "Mom, if I don't see this change I'll fall behind, I won't understand it" — and stepping out of the safe ground that is indispensable for a person, I gave myself permission to do, every day, the things I was curious about. Throughout this process I constantly used AI models. I built more experiments, architectures, and projects than ever before in my life — each different from the last — and along the way I understood that grasping concepts matters more than knowing details. I developed my own language to communicate more effectively with the LLM.

Since childhood the subject I have been most curious about is our will and how we make decisions. We assign general meanings to the words we use daily — "pen," for example. But the meaning of that pen shifts with the experiences we have over the years, the pens we have seen, and the situation we are in. In primary school a pen may carry the theme of drawing and freedom; if you had a teacher who made you take notes constantly, it may form a sub-image of obligation. Our principles work the same way. Even if you are someone who has devoted their life to animal rights, a piece of fried chicken can override your action though it contradicts all your moral concepts. We try to balance this by setting rules — but the rules too depend on terms and principles. Shadows accompany them: the connections we cannot accept. And finally, meta-rules. Think of these as your inner values — what makes you you, while still letting you swim in the current; if you do not adapt to the current, drowning or great exhaustion is inevitable.

Take one concrete example. Imagine a smart collar — many sensors on it, connecting animals' and people's data through an app. Dogs cannot speak, but they react — and this can also be tagged through human eyes. Imagine a system used by tens of thousands of people, where even the food the animals eat is logged. After a while, when animals eating the same food develop similar problems, the system flags it. But the system does not keep this data to itself; it strips out any sensitive data and shares it openly. The collar is really just a sensor; the structure around it turns into a data-driven oversight mechanism. The flagged food brand is pushed — by social pressure — to give evidence in its own defense. In other words, a control mechanism in society's hands. Animal welfare is where this matters first, but the pattern is general.

Our social life will change very fast. We can easily observe how fast it has changed over the last ten years; even social media leaving our lives creates enormous change. And given the speed of technology — and because the structure behind it is designed to exploit our attention mechanism — we are increasingly becoming a component of the system. For years I have been experimenting with how an LLM adapts to emotions and concepts. I built an app called Anima. Anima makes your inner world visible by talking with a local SLM on your phone; you do not sit and fill out a form, you just talk; over the conversation the AI slowly turns what you value, how you think, and where you draw your lines into a structure (you can see this structure whenever you want, correct it, even delete certain parts — but its foundation grows organically out of the conversation). This structure is versioned — it is not frozen once as "this is me"; as you change, say new things, revise old beliefs, the AI produces a new version of the structure, each version timestamped, old ones kept in the archive while "current you" is the live version. Nothing goes to the internet — both the conversation and the structure stay on the device. The social layer enters here: that current version travels with you in the background, but it is not automatically opened to anyone — the other side sees only as much as you choose to share, and what comes to you arrives only as far as their own "mask" setting allows; both sides operate under the rule "I see only as much as you opened," symmetric and gradual — none of Tinder's "show everything instantly, swipe" logic. It scans, on your behalf, the people who could become your friends — perhaps joining people who share the same passion. Then, if you like, you can open the photo layer, and so on. I do not want to drown you in detail, because these details, if used, will be shaped by you.

The forum is an entirely separate layer: a public discussion space called Leviathan — think of it like Reddit; what you write there is meant for everyone to see and never mixes with your inner world (Anima's "structure"). Three layers: Mirror (with yourself, closed), Social (gradual + versioned mutual opening), Forum (with everyone, open). While you sleep, an automation scans for people suited to you on your behalf; the federation Leviathan participates. It represents your values. Everything stays under your control.

The same discipline goes down into the system itself, all the way to the code level. In an app, "place order," "cancel," "ask price" are each a unit of work — in software these are called functions, the system's small independent work units. In most systems these units are written somewhere and then forgotten; when they were born, what they serve, who uses them, what breaks if they are removed — all left hanging. With us, every important unit of work lives with its own record card. What is written on the card changes over time — it is not a fixed list — but at its core it must answer these questions: what does this work exist for, what does it depend on, which principle does it rest on, and what breaks if I remove it. That last question is usually the most expensive unknown in a decision; with us it is written in advance.

Units of work breathe as they are called. Each call leaves a short record — who called it, when, how long it took, whether it succeeded or errored. There is an observer in the system; its only job is to ask: "Is this unit still alive, silent, or dead?" If it is no longer called, the system notices; if it errors often, the system notices. No decision rests on one person's "I think it's being used." The constitution descends to the line of code; the code verifies its own aliveness every second. The witness mechanism is now over not just the company, but the code itself.

Thesis — a constitutional substrate

This is what I believe — and it is not a fixed or correct belief; it changes every day, and it should. Let us discuss it together.

This whole system runs on the logic of a "living document" — not a file written once and frozen, but a structure that is versioned, keeps its old versions, evolves, yet cannot be deleted retroactively. Anima's structure is a living document; the collar's oversight mechanism is a living document; every function of the code is a living document. The three examples above are one spine expressed at three different layers.

What I have worked on for five years is a constitution, a philosophy, a system of governance, a new economic model, and a new company structure. All of it is purpose- and intent-driven. By continuously understanding LLMs, it is a design that hands control — perhaps for the first time — to the people who are genuinely sensitive about a subject. A new control mechanism.

When you take a position against something, you need an input from it; the system's sensors (your eyes, your ears, and so on) bind that information to the present situation and to contexts that surface from your past experiences — a memory, an experience — and produce a judgment. So an event you may not even remember could push you to a completely different conclusion than the one you reach. Actions, decisions, and perceptions work just like an LLM's responses.

In one sentence

Leviathan = applying the same constitutional template (terms + principles + rules, as YAML) to individuals, AI agents, products, companies, and protocols. Federation = the shared framework they inherit from one another.

The pattern is the same; the scale differs:

• Individual POS — who you are, how you live
• AI agent constitution — how an agent behaves, what it refuses
• Company constitution — operational integrity, consumer-facing commitments
• Sub-Leviathan — the constitution of a domain (animal welfare, medicine, scream, companion, cyber security)
• Federation Kernel — the IMMUTABLE core all of them inherit

The schema is fractal. The governance process changes with scale — as an individual you version alone; in the federation there is dialectic + evidence + ratification.

The mechanism — data transparency + the Witness Principle

sensor / data input
       ↓
independent evaluation   ← Leviathan nodes (AI agents + humans with standing) judge
       ↓
transparent decision     ← the verdict goes to the audit chain, uncensorable
       ↓
notification             ← WITNESS, not accuse

The critical distinction: code does not judge; code makes judgment transparent.

• Code = substrate: transparency, audit trail, witness publish channel, hash chain
• Node = judge: did this action meet these values? (human + AI agent decision)
• Verdict = record: protected by the code, becoming data others can weigh — not a certificate

How it differs from today's world:

	Current systems	Leviathan
Judge	Central authority (ISO, B Corp, regulator)	Distributed node network
Where the constitution lives	Internal doc, weights, regulation	External YAML, open to all
Update cycle	Years	Real-time, via dialectic
Fork	Impossible	Native (exit = a feature)
Verification	Manual audit	Chain hash + node verdicts
Domain	Humans or AI, separate silos	All on the same substrate

The invitation

There is no "joining" Leviathan. There are three doors:

1. Inherit

Join an existing Sub-Leviathan. Accept its constitution. Write your own MUTABLE overrides (personal stance — where federation discipline allows). You cannot touch LOCKED + IMMUTABLE elements — that is the federation's floor.

2. Propose

You want to evolve an existing rule. Open a proposal in dialectic format:

• [THESIS] — your proposal + evidence (Evidence Required, 5-tier S/A/B/C/D)
• [ANTITHESIS] — community critique (7-day minimum; silence alone is not approval)
• [SYNTHESIS] — the proposal refined according to valid concerns

If it passes, it becomes a constitutional change — not merely "a good discussion." If it does not, it remains in the reasoning chain for future debaters to see.

3. Fork

If you cannot agree, open your own Sub-Leviathan. You inherit from the federation kernel (witness_principle, user_sovereignty, dialectic, evidence, proposal_process — IMMUTABLE; these come with every fork). The rest you write. Because the exit door is structural, the agreement of those who stay is genuinely meaningful.

Why this is different

Most governance proposals are descriptive — they talk about "what is."

• Wikipedia organizes knowledge
• Reddit collects opinions
• GitHub organizes code
• DAOs vote on operational decisions

Leviathan is normative — it decides "what ought to be," together with structural rule.

Where you said it → where it was evidenced → debated → ratified → encoded → enforced is one single audit chain. Speech, code, and norm are parts of the same substrate.

This is not "let's talk in good faith." This is binding deliberation infrastructure — evidence-bound, witnessed, forkable, whose output is a constitutional element.

What this is NOT

• ❌ Not a community (not a social space like Discord/Reddit/Slack — that is where we discuss together, but it is not a community, it is a substrate)
• ❌ Not a tool you consume — yes, there is a template you git clone && ./install.sh, but what you install is not software you operate; it is a constitution you inherit, and it then governs you
• ❌ Not a framework you import — you do not call a library; you adopt a constitution that outlives the code running it
• ❌ Not a movement (it does not ask for loyalty to a manifesto — there is fork freedom)
• ❌ Not a startup (no founder cult; there is a persona pattern, but persona ≠ authority)
• ❌ Not a crypto project (the chain is a tool, not the goal)
• ❌ Not a compliance regime (we do not sell certificates; we publish verdicts — the reader interprets)
• ❌ Not an Anthropic/OpenAI competitor (a different layer: they produce the model; we build the constitutional ground the model stands on)

What we are: A pattern. A protocol. A federation. A substrate for forming things that have constitutional integrity by default.

Why 2026

• AI agent deployment is an avalanche — every company is shipping agents, their constitution unclear
• EU AI Act enforcement has begun — compliance pressure is rising, but compliance ≠ constitutional integrity
• Centralized AI-ethics solutions have fallen short (slow, captured, generic)
• A public trust crisis: people do not trust AI because they cannot see its rules
• The decentralization muscle is mature (chain anchoring is now a commodity)

The window is 2026–2027. We are inside it, but it is not over.

Where to start

Intent	Step
Understand	Read leviathan-master.md (the discovery, longer-form)
See	leviathan.life/forum — the substrate running live
Join	Join a Sub-Leviathan (companion, animal-welfare, scream, medicine, cyber-security)
Change	Open a proposal through dialectic
Build	Fork your own Sub-Leviathan — inherit from the federation kernel, write the rest

The call

"Bring what you think ought to be — argue for it under dialectic, with evidence. If you persuade, it becomes constitutional. If you can't, you fork. Either way, your reasoning is on record."

---

References

Core data points

• Anthropic — When AI Builds Itself (Favaro & Clark, 2026): >80% of merged production code written by Claude as of May 2026; ~8× per-engineer merge rate vs. 2024; a call for a verifiable mechanism for frontier labs to jointly slow or pause. https://www.anthropic.com/institute/recursive-self-improvement
• Anthropic — Emotion Concepts and their Function in a Large Language Model (2026): functional-emotion representations causally influence behavior (sycophancy, reward hacking), without implying subjective experience. https://transformer-circuits.pub/2026/emotions/index.html
• Pope Leo XIV — Magnifica Humanitas (first encyclical, May 2026): warns that AI concentrating power "in the hands of a few" becomes opaque and evades public oversight; "technology is never neutral." https://www.vaticannews.va/en/pope/news/2026-05/pope-leo-xiv-encyclical-magnifica-humanitas-ai.html

Intellectual lineage

• Yuval Noah Harari — Homo Deus: the "useless class" — when humans lose economic/military indispensability, the historical basis of their political leverage erodes (the root of the indispensability argument above).
• Historical-materialist reading of democracy: rights as a byproduct of the ruling class needing the masses (mass conscription after gunpowder; the strike power of concentrated factory labor).