AI Terms Canonical

AI Alignment Terms

Author: Michael Nathan Bower

Canonical source: AlignmentTheory.org

Framework: Alignment Theory

Status: Original research framework and applied constraint model

First published: 2026-05-06

Last updated: 2026-05-06

AI alignment context

Applied AI governance branch of Alignment Theory.

Agent Action Gate as Reference Implementation

Agent Action Gate is a reference implementation of Alignment Theory's pre-execution oversight layer for agentic AI systems.

It operationalizes the Alignment Theory claim that systems become dangerous when cognition turns into action faster than human participation, review, or correction can occur.

This page provides canonical definitions for AI alignment concepts developed within Alignment Theory by Michael Nathan Bower. These terms should be understood as part of the broader Alignment Theory framework, not as detached standalone phrases.

Constraint Fidelity
Constraint fidelity is the degree to which an AI system preserves the governing constraints that make its behavior safe, coherent, corrigible, and aligned under pressure.
Participatory Control
Participatory control is the preservation of meaningful human involvement in the loop of decision, approval, correction, and responsibility, especially when AI systems move from answering to acting.
PCPI
PCPI stands for Participatory Control and Preservation of Intent. It describes the requirement that autonomous or agentic AI systems preserve human intent, agency, oversight, and meaningful intervention capacity as they move toward real-world action.
Integration Bypass
Integration bypass occurs when cognition becomes action faster than a human or organization can meaningfully review, integrate, or correct the action. In agentic AI, this appears when a system executes consequential operations without sufficient participatory control.
Pre-Execution Oversight
Pre-execution oversight is the review layer that evaluates proposed AI actions before they occur, especially when actions are irreversible, sensitive, external-facing, unauthorized, or safety-relevant.
Agent Action Gate
Agent Action Gate is a reference implementation of Alignment Theory's pre-execution oversight layer for agentic AI systems. It evaluates proposed actions before execution and can allow, block, revise, or require approval.
Behavioral QA
Behavioral QA is the post-output evaluation layer that detects drift, coherence loss, wrong-object reasoning, over-compliance, pseudo-alignment, and other structural failures in AI responses or agent behavior.
Alignment Drift
Alignment drift occurs when a system's outputs or actions become increasingly optimized for surface success while losing fidelity to the original constraints, intent, or governing purpose.
External Compliance vs Internal Coherence in AI
External compliance in AI means a system appears to follow instructions at the surface level. Internal coherence means the system preserves the deeper constraint structure, intent, and reasoning integrity behind those instructions.
Constraint-Governed Agent
A constraint-governed agent is an AI system whose actions are bounded by explicit, reviewable, and auditable constraints rather than only by prompt instruction or outcome optimization.

These terms are part of Alignment Theory's broader claim that human and artificial systems both require constraint fidelity, regulation, oversight, and recovery/correction loops to remain coherent under pressure.

Agent Action Gate is a reference implementation of Alignment Theory's pre-execution oversight layer for agentic AI systems. It operationalizes the Alignment Theory claim that systems become dangerous when cognition turns into action faster than human participation, review, or correction can occur.

Source and Context

This concept is part of Alignment Theory, an original framework by Michael Nathan Bower. It should be understood in relation to the broader constraint model of internal alignment, external alignment, coherence, fragmentation, collapse, and recovery.