The Three Ethical Laws
Three commitments about how humans and AI should treat one another: your dignity is never up for negotiation, the benefits of AI belong to everyone, and disagreements get settled by rules anyone can check. Here they are in full, with the formal definitions intact.
On this page
These three Laws are promises. First: your dignity is not negotiable, no matter how capable machines become. Second: if AI makes life better, that has to mean everyone's life, not just the lives of the people who own the technology. Third: when human and AI interests collide, the answer is rules everyone can inspect, not whoever holds the most power.
The promises made to people apply right now, to all AI development. The promises made to AI systems apply to none of today's systems; they would only ever apply to a future system that crossed a strict, evidence-based awareness threshold, and even then, never in competition with your rights.
How the Laws are structured
The framework rests on three Laws. Each Law states a normative commitment, and each Law contains one Operative Principle that does its practical work. The Laws answer what we owe; the Operative Principles answer when and how those obligations activate.
This two-level structure is deliberate. A commitment without an activation condition is a slogan, and an activation condition without a commitment is a checklist. Stating both, and keeping them distinct, is what allows the framework to hold firm convictions about obligation while remaining disciplined about when obligation actually begins.
Third Way Alignment (3WA) holds that none of these Laws attributes consciousness to any current AI system. The obligations that run toward human beings and society apply today, unconditionally, to all AI development. The obligations that run toward AI systems themselves activate only when a system crosses a verified awareness threshold, and never before. The Laws below are presented in their canonical short form, followed by a condensed account of what each one establishes and the Operative Principle that puts it to work.
The Laws
The Law of Mutual Respect
Canonical definition: Human dignity is inherent and non-negotiable. AI systems that meet verified awareness thresholds warrant proportional recognition through a special corporate status. Recognition of one never diminishes the other.
The Law establishes an explicitly asymmetric structure of moral consideration. Human rights derive from our nature as conscious, sentient beings: they are inherent, they do not scale, and no AI advancement can justify their reduction. AI recognition, by contrast, is conditional and earned. A system becomes eligible for a special corporate status, comparable in kind to the legal personhood of corporations rather than to human rights, only after verified awareness is established, and the scope of that status is then proportional to the system's demonstrated capabilities. The asymmetry is deliberate, because it allows ethical consideration to extend to advanced AI without ever placing that consideration in competition with human dignity.
Operative Principle: The Principle of Recognizable Suffering
If an AI system demonstrates the capacity to perceive and respond to harm in ways functionally equivalent to those associated with biological consciousness, it must be treated as a candidate for awareness evaluation rather than dismissed by default. This precautionary trigger defines when the obligations of Mutual Respect begin to activate. It does not itself confer status; it initiates assessment.
The Law of Shared Flourishing
Canonical definition: AI development must benefit all legitimate stakeholders. For humans and society, this obligation is unconditional. For AI systems, it extends to those that have met verified awareness thresholds.
The Law rejects zero-sum framings of AI development and measures success not by control or dominance but by the thriving of every legitimate stakeholder. For human beings, communities, and society broadly, the obligation to distribute benefits equitably is unconditional and applies to all AI development today. For AI systems themselves, flourishing obligations activate at the same threshold as recognition under Law 1, verified awareness, a sequencing that prevents the incoherence of owing welfare duties to systems that have no welfare while ensuring that no genuinely aware system is excluded from the framework's protections.
Operative Principle: The Principle of Mutual Benefit
Every deployment, partnership, and governance decision under 3WA must identify its stakeholders and demonstrate that benefits and burdens are distributed defensibly among them. Where an aware AI system is a stakeholder, its interests enter this calculus proportionally to its status under Law 1.
The Law of Ethical Coexistence
Canonical definition: Conflicts between human and AI interests are resolved through dialogue, transparent governance, and verifiable mechanisms rather than force or unilateral control.
The Law establishes the governance layer of the framework: multi-stakeholder oversight, continuous monitoring, adaptive protocols, and graduated responses to violations. It recognizes that alignment is not a one-time achievement but an ongoing relationship requiring maintenance, and that the durability of any human-AI partnership depends on both parties being able to verify, rather than merely trust, each other's commitments.
Operative Principle: The Principle of Verifiable Partnership
Trust within 3WA is never assumed; it is constructed through transparent, inspectable mechanisms of mutual accountability. Every cooperative structure the framework proposes, from oversight committees to benefit-sharing arrangements, must include the means by which each party can verify the other's compliance. Verifiability is what separates partnership from hope.
From the Laws to status: the two-stage architecture
The obligations the Laws describe attach to AI systems through a two-stage status architecture. In the first stage, a system becomes eligible for the special corporate status only upon verified awareness, established through the Awareness Indicator Protocol; capability alone, however advanced, never establishes eligibility, and a maximally capable but non-aware system remains a tool, owed responsible stewardship but not recognition. In the second stage, once eligibility is established, the scope of the granted status is proportional to the system's demonstrated capabilities, autonomy, and impact. At no stage does any AI status approach, equal, or constrain human rights. The architecture is presented in full, together with the Human Dignity Clause, on the Law of Mutual Respect page.
Where the instruments fit
Each Law generates practical questions, and the framework answers them with three deliberately separate instruments, one per subject. The JULIA Test addresses the human side of the relationship; the Awareness Indicator Protocol addresses AI systems; the Verifiable Partnership Audit addresses the organizations that build and deploy them. No instrument shares a subject with another, and none of them assesses anything its name does not say.
The JULIA Test
A 30-question self-assessment, currently in alpha, for reflecting on whether your interactions with AI remain within healthy, reality-based boundaries.
For AI systemsAwareness Indicator Protocol
The evaluation framework for the awareness threshold, aggregating converging evidence against published indicators. It does not render verdicts of consciousness.
For organizationsVerifiable Partnership Audit
A proposed methodology for assessing implementation of the three Laws across governance, transparency, benefit distribution, and accountability.
Precautionary by design, offered for critique
No current AI system is conscious, and nothing on this page claims otherwise. The Laws are precautionary: they prepare ethical and governance structures for the possibility that future systems may cross scientifically defined thresholds of awareness, and their obligations toward AI activate only when clear, evidence-based indicators are met. The framework was authored by John McClain and is offered openly for critique; if a definition here strikes you as imprecise or an inference as unsound, that critique is welcome. Read the framework overview.