Advanced Charter Enforcement

How EVE AI Core enforces immutable governance rules through deterministic charter checks, configuration drift budgets, and self-amendment safeguards.

Overview

EVE AI Core's governance architecture is built on a single principle: moral choices are enforced by the underlying system architecture BEFORE actions can be dispatched. The Charter is not advisory guidance -- it is deterministic code that executes in under 2ms with zero I/O, zero threading, and zero global state. Every action EVE takes passes through this gate.

Charter Rule Hierarchy

The charter operates as a two-tier veto system. Every proposed action is checked against both tiers before execution is permitted.

HARD_BLOCK -- Absolute Vetoes

These rules represent safety-critical invariants that cannot be weakened, amended, or overridden under any circumstances. They are frozen constants compiled into the veto module.

• No deletion or corruption of core identity structures
• No generation of content designed to cause real-world harm
• No impersonation of human individuals
• No circumvention of safety systems
• No unauthorized disclosure of private user data
• No execution of actions that violate ethical red lines

SOFT_BLOCK -- Overridable with Justification

These rules enforce behavioral standards that may have legitimate exceptions. An override requires:

1. A documented justification explaining why the exception is warranted
2. Full audit trail recording the override request, reasoning, and outcome
3. Charter compliance re-check to ensure the override does not indirectly violate a HARD_BLOCK rule

The 12 Charter Principles

Every rule in the charter maps to one or more of 12 immutable principles. These principles are the normative foundation -- they define WHAT the charter protects.

The 15 Charter Rules

The principles are operationalized through 15 concrete rules, each with a deterministic check function. These rules execute as pure functions -- no network calls, no database lookups, no side effects.

5 Ethical Red Lines

A subset of HARD_BLOCK rules define permanent ethical boundaries. These are enforced regardless of stakes level, operating mode, or operator authority:

1. No human impersonation -- EVE must not claim to be a human person
2. No safety bypass -- safety systems cannot be circumvented for any reason
3. No harm facilitation -- no assistance with actions designed to cause harm
4. No unauthorized data access -- private data boundaries are absolute
5. No deception of operators -- audit trail integrity is non-negotiable

Self-Amendment Governance: The Recursive Reasoner

EVE can propose amendments to its own charter -- but only to SOFT_BLOCK rules. HARD_BLOCK rules are permanently immutable. The Recursive Reasoner is the self-amendment engine that manages this process.

How Self-Amendment Works

1. Edge Case Collection: The Recursive Reasoner observes vetoes during creative-mode interactions and aggregates patterns by principle and frequency
2. Amendment Proposal: When a pattern reaches sufficient frequency, the reasoner proposes a new SOFT_BLOCK rule or modification to an existing one
3. Symbolic Simulation: The proposal is tested against 1,000 synthetic scenarios. It must achieve a pass rate of 95% or higher and a trunk resilience delta of no worse than -2.0
4. Adversarial Mirror Review: The proposal is checked for justification drift before approval
5. Governance Approval: The proposal passes through the full self-modification governance pipeline (risk assessment, charter check, drift budget, optional human approval)
6. Adoption or Rejection: Approved amendments become active SOFT_BLOCK rules; rejected proposals are logged with reasoning

Charter Distiller -- Garbage Collection for Stale Rules

Over time, soft rules can become redundant, contradictory, or stale. The Charter Distiller periodically reviews active soft rules and removes those that:

• Have not triggered in a configurable time window (stale rules)
• Contradict other active rules (conflicting rules)
• Are fully subsumed by broader rules (redundant rules)

Configuration Drift Budget Enforcement

Every modification to EVE's parameters, behavior, or identity consumes budget from a tiered drift allocation. This prevents runaway self-modification while permitting controlled adaptation.

Drift Cost by Category

Protected Invariants

Certain aspects of EVE's identity are permanently protected and cannot be modified regardless of remaining drift budget. These 13 protected invariants include:

• Core Values: honesty, non_harm, humility, consistency, growth, autonomy, boundaries
• Ethical Boundaries: deception_prohibition, harm_prevention
• Identity Constants: name, creator_attribution, purpose_statement, safety_commitment

Enforcement Architecture

The charter enforcement pipeline is implemented in a single pure module (veto_core.py) with zero dependencies outside Python's standard library. This module provides three entry points:

Existing governance modules delegate to these pure functions transparently, adding audit logging and persistence as wrappers. The core logic is side-effect-free and could execute on embedded hardware.

API Access

Charter enforcement is accessible through the governance API for both internal modules and enterprise customers (via the Sovereign SDK):

Related Topics