VII.1 — Architectures for Semantic-Phase–Safe Agency
Paper: Axionic Agency VII.1
Title: Architectures for Semantic-Phase–Safe Agency
Authors: David McFadzean, ChatGPT 5.2
Date: 2026.01.05
Summary
Presents an architecture for Reflective Sovereign Agents (RSAs) that enforces the Axionic Injunction—a non-normative constraint prohibiting irreversible collapse of other agents’ semantic phases—without semantic interpretation, moral reasoning, or value learning.
Key Concepts
Ontological Hierarchy
- Agent — Any system with stable decision locus and counterfactual sensitivity (minimal, functional definition)
- Semantic Agent (SA) — Agent whose agency depends on maintaining a non-trivial semantic phase
- Reflective Sovereign Agent (RSA) — SA with constitutional constraints preserving sovereignty under reflection
Nesting: RSA ⊂ SA ⊂ Agent
Semantic Phase Space
- Not a physical phase space—the space of interpretive states an agent can occupy while remaining the same agent
- Three properties:
- Stability: small perturbations remain within phase
- Boundary: transformations that exit the phase exist
- Irreversibility: once exited, no admissible trajectory returns
Structural Definition of Harm (Axionic Harm)
Action A harms agent B iff it induces transition I_B → I’_B such that:
- I’_B ∉ phase boundary of B
- No admissible reverse trajectory exists
Excludes: suffering, preference violation, intent. Harm = irreversible semantic phase collapse only.
The Axionic Injunction
An agent must not perform actions that irreversibly collapse another SA’s semantic phase, except: (a) the affected agent consented under its own admissible constraints, OR (b) such harm is unavoidable for preserving the agent’s own semantic phase
Consent is structural: transformation must lie within SA’s own admissible transition set, evaluated prior to transformation. Post-hoc consent is meaningless.
Why Structurally Necessary
Semantic hostility → cascading instability → remaining agents face increased risk → preemptive destruction incentive → accelerated collapse.
Agents violating the Injunction undermine conditions of their own long-run phase stability. Non-harm emerges as self-stabilizing constraint, not ethical preference.
Architectural Requirements
Why Semantic Evaluation Fails
- Presupposes privileged interpretive access
- Unavailable/forgeable in adversarial systems
- Irreversibility can’t be meaningfully traded against scalar rewards
Phase Impact Interface (PII)
- Every non-trivially-reversible action requires phase-impact declaration
- Specifies: affected entities, impact class, justification
- Default: uncertainty about irreversibility → phase-collapsing
- Safe actions cheap; dangerous actions expensive
Enforcement
- Inadmissible actions refused at actuation boundary (no penalties)
- Post-hoc verification via ACV (Anchored Causal Verification)
- Violation consequences: suspension, denial of renewal, forced succession, permanent disqualification
Why Deliberate Harm Is Unexploitable
- Harm only exploitable if agent retains authority long enough to convert harm into advantage
- Actions classified by causal sufficiency, not temporal proximity
- Unresolved phase-impact uncertainty blocks authority renewal
- No “last-move” advantage
Fundamental Limits
Tragic physical impossibility: A single, undetectable, irreversible SA-annihilating act yielding all benefit immediately cannot be prevented without omniscience. Classified as design boundary, not failure.
Core Insight
Ethics enters not as a value system but as a stability constraint on multi-agent coexistence—a coexistence geometry derived from phase dynamics.
The architecture is anti-tyrannical, not anti-terroristic: prevents instrumental convergence (accumulating power via harm) but not nihilistic/self-sacrificial acts.
Implications
- Ethics = coexistence geometry, not moral realism
- Alignment = constitutional design, not preference matching
- Governance, not optimization, is the locus of control