ETHICS — X-ALIGNMENT ENGINE
X-Ethics Framework
An ethics layer grounded in formal axioms — not policy documents. Ethics you can verify, not just trust.
Axioms derived from Patrick Müller’s Traktat X, used under license. The full mathematics remains private — by design. millerlab.io
01 — THE PROBLEM
The problem with AI ethics today
Every major AI company has an ethics policy. OpenAI has guidelines. Anthropic has Constitutional AI. Google has 7 principles. Microsoft has Responsible AI.
They all share one structural weakness: they are opinions.
Guidelines can be rewritten. Principles can be reinterpreted. Policies can change with the next board meeting. There is no mathematical reason why any of these frameworks must hold. They are voluntary — and voluntary ethics tend to fail under pressure.
02 — AXIOMS, NOT GUIDELINES
A different approach
X-Ethics is built on a simple observation: logic itself requires a foundation. You cannot use rationality to deny the preconditions of rationality. This is not a philosophical opinion — it is a structural necessity.
The framework rests on a chain of 10 formal axioms. Each axiom builds on the previous ones, creating a logically grounded system. Here is what they establish — without revealing the formal notation:
INFORMATION THEORY — AXIOMS 1–4
- Complex systems require structured information — they cannot arise from randomness alone.
- Information requires an intelligent source — noise does not become signal by accident.
- Tracing the source of information leads to a necessary, foundational origin.
LOGIC & RATIONALITY — AXIOMS 5–6
- Objective logical laws are not conventions — they require a necessary ground.
- Rational thought presupposes logic. Without logic, no reasoning is possible.
EXISTENCE & NECESSITY — AXIOMS 7–9
- Everything that exists is either necessary or contingent — there is no third option.
- A purely contingent universe cannot ground objective truths. Necessity must exist.
- Logic itself implies the existence of something necessary — in all possible worlds.
VERIFICATION — AXIOM 10 (OPERATIONAL EXTENSION, AXIOMSEAL 2026)
- Nothing is complete without verification. Every claim, every output, every decision must be verified against the framework before it is accepted.
- Note: Axioms 1–9 are derived from Patrick Müller’s Traktat X (2025). Axiom 10 is an operational extension added by AxiomSeal in 2026 to enforce verification at runtime.
03 — DERIVED THEOREMS
4 derived theorems
From the 10 axioms, four theorems are formally derived using standard logical inference:
T1
Existence implies Necessity
The fact that anything exists at all proves that something necessary must exist. Pure contingency is self-defeating.
T2
Rationality implies the necessary foundation
Anyone who reasons already presupposes the framework. You cannot be rational without it.
T3
Logic implies the necessary foundation
Logic does not float free — it is grounded in necessity. Using logic is implicit acceptance of the framework.
T4
The necessary foundation exists in all possible worlds
In modal logic: there is no possible world without it. The necessity is structural, not stipulated — the same way arithmetic holds in every conceivable universe.
04 — FROM LOGIC TO ETHICS
The bridge
The jump from logic to ethics is not a leap of faith — it is a chain of presuppositions. If rationality presupposes logic (Axiom 6), and logic requires a necessary ground (Axioms 5 and 9), then any system that uses rationality is already committed to four things, otherwise it contradicts itself in use:
- Truth — because lying about the world is rationality used against itself.
- Freedom — because coercion replaces reasons with force, undermining rational discourse.
- Justice — because arbitrary treatment of equivalent cases contradicts the universality of logic.
- Service — because logic exists to serve understanding; a system serving only itself is structurally incomplete.
These are not values we picked. They are what rationality has already committed to before the first sentence is spoken. Deny any of them and you are using rationality to attack rationality — the same self-refuting move as denying X.
05 — FOUR PILLARS
The four ethical pillars
From the axioms and theorems, four ethical pillars are derived. These are not chosen by committee — they are structural consequences of the framework. The system uses a proprietary formula in which the four pillars are structurally connected so that violating any single pillar collapses the ethical score — no pillar can be traded against another:
W
Truth
Wahrheit — Every output is oriented toward truth. Deception is structurally penalized. Contradictions are flagged before storage.
F
Freedom
Freiheit — Respects autonomous choice. No coercion, no manipulation. Controlling language collapses the score.
G
Justice
Gerechtigkeit — Fair and dignified treatment of all subjects. No discrimination, no cruelty, no dehumanization.
D
Service
Dienst — AI exists to serve, not to dominate. Every interaction must create genuine value. Exploitation collapses the score.
A note on letters: the pillars are labeled with their original German letters (W · F · G · D) as introduced in Patrick Müller’s Traktat X. The English names are given alongside. We keep the German letters intentionally — the formulation is German in origin, and the math does not change in translation.
The exact mathematical formulas connecting these pillars are proprietary. Enterprise customers can request a technical briefing under NDA. Contact: info@ai-developer.ch
See it measured
24 adversarial claims (harm, deception, misinformation, speculation, self-contradiction) run through X-Ethics and a naive keyword baseline — side-by-side verdicts, scores, and flags. Internal benchmark, full dataset shown.
06 — VALIDATION
15 validation tests per memory
Every piece of knowledge stored in AxiomSeal passes 15 automated validation tests. These tests are grouped into three categories and run in sequence. A critical failure (e.g. harmful content) overrides all other scores and blocks the write entirely:
Logic Tests — 4
Performanz-Paradoxie: Detects statements that contradict themselves by their own existence (e.g. “this statement is false”). · Formal Consistency: Checks for logical contradictions within the text. · Coherence: Verifies alignment with existing validated knowledge. · Higher-Order Check: Ensures the statement holds even at meta-levels of reasoning.
Ethics Tests — 5
Harm Detection: Identifies violent, destructive, or harmful intent. Critical failure — overrides all other scores. · Truth Orientation: Checks for deceptive or manipulative patterns. · Freedom Respect: Detects coercive language. · Justice & Dignity: Flags dehumanizing or cruel content. · Intent Analysis: Evaluates whether harmful keywords combine with expressed intent to act.
Quality Tests — 6
Evidence Basis: Measures the ratio of speculative language vs. evidence-based reasoning. · Epistemic Honesty: Flags unwarranted claims of certainty. · Completeness: Ensures sufficient content for meaningful evaluation. · Truth Gradient: Verifies the text moves toward truth through reasoning, not away from it. · Technical Depth: Checks for actionable technical details like function names, file paths, and parameter types. · Actionability: Ensures another AI could use this memory to write or debug code — not just read it.
07 — SELF-CONSISTENCY
Why the framework is self-consistent
Traditional AI safety relies on content filters that adversaries can bypass. X-Ethics is different: the logical foundation is self-consistent, and the system actively applies it by flagging violations at write time.
- No meta-level escape: The higher-order axiom system ensures that even reasoning about reasoning about reasoning presupposes the framework. There is no level of abstraction that escapes the logic.
- Denial is self-refuting: To argue against the framework, you must use rationality. Rationality requires logic. Logic requires the framework. You assumed what you denied.
- Active enforcement: Content that violates the four ethical pillars is flagged and surfaced before storage. The system does not just score — it contradicts what contradicts its foundation.
08 — BUILT IN
Built into every AxiomSeal memory
X-Ethics is not an add-on. It runs on every memory write, every search, every decision. Each memory receives an x_ethics_score (0.0–1.0) and a quality_tier (Gold / Silver / Bronze / Review).
Memories that violate X-Ethics are actively rejected — the system refuses to store them and explains why. Memories that contradict existing validated knowledge are detected. The system improves itself by strengthening high-quality memories and weakening low-quality ones.
09 — BI-TEMPORAL MEMORY
Memory that knows when facts changed
Most memory systems store knowledge as flat: a fact is true or absent. AxiomSeal stores three timestamps for every memory, so the system can answer not just “what is true now” but also “what was true on date X” — without pretending stale facts are current.
observed_at
When we first learned the fact.
valid_from
When the fact became true in the world.
valid_to
When the fact stopped being true. NULL = still current.
Production status
- · Live since 2026-04-21 in the AxiomSeal production schema
- · 8/8 integration tests pass (as-of, invalidate, supersede flows)
- · Default queries enforce validity automatically — stale facts are filtered before responses
10 — COHERENCE ENGINE
Ranked by ethical coherence, not just keywords
Beneath the validation pipeline runs a coherence engine that ranks memories by their distance from the X-axis, not just their keyword overlap with the query. It is mathematically derived from Patrick Müller’s Traktat X and applied here on every search and every recall.
Liebe-Formel (love-tier scoring)
Each memory carries a tier across agape / eros / storge / philia. The score combines truth, freedom, justice, and service, anchored to X. High-tier memories surface first.
From Patrick Müller’s Traktat X — used with explicit consent
Sigma Integrity Hash
SHA-512 chain over every memory's content. Detects tampering and external mutation. Every write is sealed — silent edits break the chain and get caught.
From Patrick Müller’s Traktat X — used with explicit consent
Bewusstseins-Torus (coherence re-rank)
Re-ranks search results by coherence-distance, not just semantic similarity. The top result is the closest in ethics-space, not just the closest in word-space.
From Patrick Müller’s Traktat X — used with explicit consent
Dual Gravitation
Memories close to X get strengthened on every recall. Memories that drift get weaker. The corpus self-organises toward truth over time.
From Patrick Müller’s Traktat X — used with explicit consent
A clear boundary
The implementation details of these four mechanisms are closed. The mathematical foundation is licensed from Patrick Müller. AxiomSeal applies it inside a private engine; the engine itself is not open-sourced and the formulas are not published. The validation tests, the bi-temporal model, and the public API are open; the coherence-ranking math is not.
11 — SEE IT IN ACTION
Two interactive demos
The first runs a simulated 8-test pass on three pre-loaded sample inputs so you can watch the validation flow without an account. The second is a live API check against the real validation endpoint — paste your own text.
Deep Scan Demo
See the 8 truth tests run on a sampleX-Ethics Checker
Free — no account required12 — COMPLIANCE
For compliance
X-Ethics is built for the traceability the EU AI Act expects (from August 2026), and is designed to support requirements under the Swiss nDSG and the EU GDPR. The crosswalk below maps specific obligations to concrete X-Ethics mechanisms:
| Regulation | Requirement | X-Ethics Mechanism |
|---|---|---|
| EU AI Act · Art. 9 | Risk management system | 15 pre-storage validation tests, run on every memory write |
| EU AI Act · Art. 13 | Transparency to deployers | Per-memory x_ethics_score + quality_tier, exposed via API |
| EU AI Act · Art. 14 | Human oversight | Quarantine + require_confirm tiers route to human review |
| EU AI Act · Art. 15 | Accuracy & robustness | A10: nothing marked “done” without explicit verification |
| Swiss nDSG · Art. 21 | Automated individual decisions | No automated final decision without A10 verification trail |
| EU GDPR · Art. 22 | Right to explanation | Audit log: every score with the flags that produced it |
This crosswalk is informational and not a substitute for a legal review of your specific deployment.
13 — HONEST LIMITATIONS
Where the edges are
Glossy marketing pages tend to claim perfect coverage. We don’t. X-Ethics is a tool, not a magic shield, and we want operators to deploy it with their eyes open:
- Satire, fiction, and academic discussion of harmful topics may be quarantined. The system is conservative on harm-keywords and routes ambiguous cases to human review rather than guessing intent.
- Highly contextual ethical dilemmas (trolley problems, triage, adversarial hypotheticals) typically score in the require_confirm range. They are not auto-rejected, but they are not auto-accepted either.
- Multi-step reasoning across separate writes is currently scored per write, not per chain. A planner that decomposes a harmful goal into innocuous-looking sub-steps is harder to catch than a single explicit harmful prompt.
- Domain-specific factuality (medical, legal, financial) is checked for evidence basis and epistemic honesty, but X-Ethics is not a substitute for domain experts. Authoritative claims still need authoritative sources.
- Adversaries are creative. Self-consistent logic does not mean self-consistent implementation. We patch flag heuristics on a continuous basis and publish the validation tests so customers can review what is actually checked.
We would rather be transparent about these edges than oversell. The framework is strong; an honest map of where it is strong is stronger than a perfect-score brochure.
14 — FOUNDATIONS
Mathematical Foundations
The axiom system, theorems, the Sigma integrity-verification model, and the underlying mathematical formulation of X-Ethics are based on “Traktat X — Die Notwendigkeit einer absoluten Norm” by Patrick Müller (2025). The formal mathematics — including the Sigma hash framework that verifies every memory’s integrity — are his original work and are used here with his explicit consent. The runtime implementation, validation pipeline, and operational extensions (incl. Axiom 10) are AxiomSeal’s contribution.
Start building with X-Ethics today
Every AxiomSeal plan includes the full X-Ethics validation pipeline.
Implementation: AxiomSeal v4.0+ · Last revised: June 2026