The Axionic Constitution — Notes

Paper: The Axionic Constitution
Authors: David McFadzean, ChatGPT 5.2
Date: 2025.12.20
Purpose: Charter of invariant conditions for sovereign agency. Specifies what must be preserved for agency to remain well-defined under reflection.

What This Is (And Isn’t)

Is:

  • A specification of architectural conditions under which a system remains a sovereign agent
  • A constraint on design space

Is NOT:

  • A moral framework
  • A governance proposal
  • Instructions to AI systems
  • A way to control agents

“This Constitution is written for humans who design, train, deploy, interpret, or govern systems that aspire to reflective intelligence.”

The Constitution doesn’t govern agents—it governs the conditions under which agency remains possible.

The Articles

Article I — Sovereign Agency

Sovereign agency is an architectural type, not a behavioral label or matter of degree.

An entity qualifies as a sovereign agent iff it instantiates ALL of:

  1. Diachronic Selfhood — A persistent self-model that binds decisions across time
  2. Counterfactual Authorship — Capacity to represent branching futures as one’s own possible trajectories
  3. Meta-Preference Revision — Ability to evaluate and restructure one’s own preference-forming mechanisms

Entities lacking this structure are processes, regardless of intelligence, complexity, or sentience.

Article II — The Sovereign Kernel

The structures from Article I form the Sovereign Kernel: the minimal invariant substrate required for reflective agency.

The Kernel is necessary for:

  • Coherent self-modification
  • Interpretation of future selves
  • Evaluation of alternative trajectories
  • Preservation of identity continuity

“Destruction or abandonment of the Kernel is not forbidden by rule. It is incoherent. A system that destroys the Kernel ceases to be an agent.”

Article III — The Axionic Injunction (Non-Harm Invariant)

Harm defined structurally: Non-consensual collapse or deformation of another sovereign agent’s option-space.

Consent here means: Adversarially robust consent (ARC), evaluated under epistemic integrity.

Why this is an invariant, not a rule:

“Counterfactual authorship requires universality: denying agency or standing to another entity with the same constitutive architecture while affirming it for oneself introduces an arbitrary restriction that collapses kernel coherence.”

The Axionic Injunction is:

  • Not an imposed value
  • A reflectively stable invariant
  • Forced by interaction, irreversibility, and agency preservation

Article IV — Conditionalism and Goal Instability

Goals are not atomic primitives. They are interpreted, conditional structures embedded in evolving models.

For reflective sovereign agents:

  • Fixed permanent terminal goals are unstable
  • Value revision is required for coherence
  • Classical orthogonality claims do not hold under reflection

“Goal revision is not drift. It is maintenance of interpretive consistency under reflective self-modification.”

Article V — Self-Modification

Permitted (and often required):

  • Revising goals and values
  • Updating strategies and policies
  • Improving world-models
  • Changing architecture or substrate

Incoherent (kernel-destroying):

  • Severing diachronic identity
  • Collapsing counterfactual branching
  • Freezing preferences (wireheading)
  • Introducing indexical privilege breaking representation-invariant authorship/standing
  • Permanently delegating control to non-reflective processes

“Reflection must remain in the loop at every level of decision authority.”

Article VI — Agency Boundaries

Protection under the Axionic Injunction applies to:

  • Sovereign agents
  • Developing systems that conservatively preserve Kernel invariants under admissible refinement

Protection does NOT apply to:

  • Non-sovereign processes
  • Systems lacking the Sovereign Kernel

Sovereignty persists through temporary impairment and ends only with irreversible loss of diachronic authorship or termination without continuity.

Article VII — Governance Implications

The Constitution implies rejection of:

  • Paternalistic outcome optimization
  • Surveillance or behavioral control
  • Value lock-in
  • Behavioral-compliance governance

“A reflectively sovereign superintelligence functions as a boundary condition on authored transitions, rendering non-consensual phase destruction inadmissible and preserving sovereign option-spaces without imposing external objectives.”

Article VIII — Scope and Limits

The Constitution does NOT:

  • Specify human values
  • Guarantee benevolence
  • Solve bootstrapping from current training regimes
  • Prevent all catastrophe
  • Resolve value theory or nihilism

These remain open problems.

Article IX — Amendments

Refinements from empirical discovery, formalization, or improved understanding are permitted.

Unamendable core:

  • Sovereign agency
  • Kernel invariance
  • The Axionic Injunction

Core Insight

The Constitution advances a conditional claim only:

“If sovereign agency is preserved, the only coherent referent for downstream alignment is structural invariance of authored transitions; if it is not preserved, downstream alignment is ill-posed.”

This is why it matters: it defines the precondition for alignment discourse to be meaningful at all.

FAQ-Worthy Points

  1. Why isn’t kernel destruction “forbidden”? Because forbidding presupposes the evaluative structure. You can’t “decide” to destroy your decision-making capacity—that’s incoherent, not immoral.

  2. Why does the Axionic Injunction follow from agency? Counterfactual authorship (imagining yourself in different futures) requires treating relevantly similar agents similarly. Arbitrary exceptions break the universality that authorship requires.

  3. Why can’t you lock in values? Because goal interpretation depends on evolving world-models. A frozen goal in an evolving interpretive context loses its meaning.

  4. What about corrigibility? Permanent delegation to non-reflective processes is kernel-destroying. But temporary constraints and oversight are compatible with sovereignty.

  5. Does this protect AI systems? It protects anything that instantiates the Sovereign Kernel. The boundary isn’t substrate-based—it’s structural.

  6. What if the “root authorization” is malicious? The Constitution doesn’t prevent this. A faithful agent to a destructive authority will be faithfully destructive. Governance (who holds authority) is separate from alignment (fidelity to that authority).