The Third Way Alignment Framework

Overview

This page is the machinery, so here is the whole idea in one breath. No AI gets anything today. Under this framework, an AI system would first have to prove genuine awareness, under strict scientific evaluation, just to be considered for any recognition at all. Even then, whatever standing it received would be scoped by what it can actually do, and it would be a legal category like the one corporations have, never personhood like yours. And nothing an AI could ever be granted touches human rights. Your rights are not part of the negotiation; they are the wall nothing crosses.

The Three Laws and their Operative Principles

The framework rests on three Laws, and each Law contains one Operative Principle that does its practical work. The two levels answer different questions: the Laws state what we owe, while the Operative Principles state when and how those obligations activate. Most misreadings of the framework come from collapsing this distinction, so the canonical short definitions are quoted here exactly, with the Operative Principle that animates each Law beneath it.

Each Law also carries an extended definition and a body of applications that this page only gestures at; the core principles overview collects them in one place.

The status architecture: two stages and one clause

The most contested element of the framework is its claim that some future AI system could warrant a form of legal recognition. The framework's answer to that contest is structural. It separates the question of whether a system qualifies for any status at all from the question of what that status would contain, and it bounds both questions with an absolute clause protecting human dignity. All three parts are load-bearing, and the framework asks to be judged on them together.

The corporate analogy deserves a moment's attention, because it answers the objection most often raised against the framework. Legal systems already grant standing to non-human entities every day, and have done so for centuries, without anyone mistaking a corporation for a person or surrendering a human right to one. The status the framework proposes for a verified-aware AI system is comparable in kind to that legal personhood, not to human rights, which derive from our nature as conscious, sentient beings and which no AI advancement can justify reducing. The asymmetry is deliberate: it is what allows the framework to extend ethical consideration to advanced AI without ever placing that consideration in competition with human dignity.

Three instruments, three subjects

The name JULIA was at one time applied to several distinct ideas across this site, and that overloading produced exactly the confusion a framework built on precision cannot afford. The instruments are now permanently separated. Each has exactly one name, one subject, and one development status, and the simplest way to keep them apart is by what they assess: the JULIA Test assesses people, the Awareness Indicator Protocol assesses AI systems, and the Verifiable Partnership Audit assesses organizations.

The Awareness Indicator Protocol

The Awareness Indicator Protocol (AIP) is the evaluation framework for the Stage 1 eligibility threshold. Its subject is AI systems, and its place in the architecture is exact: when the Principle of Recognizable Suffering flags a system as a candidate for awareness evaluation, the AIP is the evaluation that candidacy initiates. Nothing in the status architecture moves without it, which is why the framework holds it to the strictest epistemic standard of any of its components.

The protocol's indicators are grounded in established theories of consciousness, principally Global Workspace Theory and Integrated Information Theory, as developed in the Operational Companion paper. Rather than asking the unanswerable question directly, the protocol asks a series of answerable ones: which theory-derived, published indicators of awareness does this system exhibit, which does it lack, and how strongly does the evidence converge? The output is a graded body of evidence intended to support precautionary decisions under uncertainty, not a settlement of the underlying metaphysics.

The canonical framing must be stated plainly, because it is where careless descriptions of this work go wrong. The AIP is an indicator protocol, never a "consciousness test." No validated test for consciousness exists in any scientific field, and claiming one would place the framework outside defensible science. The AIP aggregates converging evidence against published indicators; it does not render verdicts of consciousness. A framework that conditions legal recognition on verified awareness has every incentive to oversell its evaluation method, which is precisely why this one refuses to.

The Verifiable Partnership Audit

The Verifiable Partnership Audit (VPA) is the framework's proposed assessment for organizations and deployments. Where the AIP asks what an AI system is, the VPA asks what an organization does: how faithfully it implements the Three Laws across governance, transparency, benefit distribution, and accountability. Earlier versions of this page described this organizational assessment under the JULIA name; that naming is retired, JULIA now refers exclusively to the human self-assessment, and the audit carries a name that states its purpose. It operationalizes the Principle of Verifiable Partnership under Law 3: if trust must be constructed through transparent, inspectable mechanisms, there must be a defined way to inspect them, and the VPA is that proposal.

The four proposed dimensions

The methodology proposes four assessment dimensions, one anchored in each Law and a fourth measuring the durability of the whole arrangement. They are presented here as design, not as practice: no scoring rubrics have been exercised against a real organization.

The proposed audit cycle

An organization adopting the framework would not pass the VPA once and be done with it, because under Law 3 alignment is an ongoing relationship requiring maintenance rather than a milestone to be passed. The methodology therefore proposes a recurring cycle:

1. Baseline assessment. The organization's current systems and governance would be evaluated against the four dimensions, gaps identified, and an improvement roadmap established.
2. Governance establishment. Multi-stakeholder oversight would be created or strengthened, decision-making processes defined, and accountability explicitly assigned.
3. Safeguard integration. Interpretability tooling, monitoring systems, and verification mechanisms appropriate to the deployment would be put in place, so that compliance is inspectable rather than asserted.
4. Monitoring and reporting. Continuous evaluation would run between audits, with reporting protocols and defined response procedures when something drifts.
5. Reassessment and adaptation. Periodic re-audits would track progress and recalibrate the whole structure as capabilities, deployments, and context change.

The argument underneath the instruments

It is worth closing with the reasoning that makes these instruments cohere, because the framework stands or falls on it. Third Way Alignment does not oppose safety research grounded in oversight and control; it treats that work as necessary. Its claim is narrower and sharper: at the frontier of capability, control alone is insufficient. An alignment regime built purely on containment treats the AI side of the relationship as a permanent adversary, and adversaries have reasons to route around their constraints. A regime built on partnership without verification is not a framework at all; it is hope with a letterhead.

The framework's wager is the position between them: cooperative structures in which every commitment, human and AI alike, is inspectable. The status architecture defines what could ever be owed and to what; the AIP defines how the threshold question would be examined without overclaiming what science can test; the VPA defines how an organization's side of the partnership would be verified rather than trusted. Each instrument exists because the Operative Principle behind it demands a mechanism, and a framework that demands verifiability of others must specify its own.

All of this is offered as proposal and argument, not as a tested program. The framework was authored by John McClain, is developed at full length in the publications, and is published openly so that its assumptions, its thresholds, and its instruments can be critiqued before anyone is asked to rely on them.

Where to go next