Third-Way Alignment: A Comprehensive Framework for AI Safety
The main thesis of the framework: the case for a cooperative middle path between controlling AI and setting it loose, and a phased plan for getting there safely.
John McClain • August 2025
On this page
This is the longest and most formal paper on the site, and the one everything else builds on. It asks a simple question: if "keep AI on a tight leash forever" and "let AI do whatever it wants" are both bad plans, what does a third option actually look like? The answer it works out is a partnership with strict preconditions, where nothing is taken on faith and human dignity is never up for negotiation. You don't need a technical background to read it, just patience for an academic format.
Access the paper
The PDF opens in your browser and can be saved for offline reading. The thesis and its operational companion are openly archived on Zenodo.
Looking for the companion paper?
This thesis argues the case; the operational guide works through the technical and ethical details of implementation.
Abstract
Third-Way Alignment (3WA) is presented as a preparatory framework for human-AI cooperation that moves beyond the binary of control versus autonomy. Writing in 2025, the thesis observes that systems of that period, such as OpenAI's o1 and Anthropic's Claude 3.5, exhibit advanced reasoning without sentience, and argues that safeguards for emerging capabilities should be prepared in advance rather than improvised after the fact. The framework is positioned as a complement to existing structures such as the NIST AI Risk Management Framework and the EU AI Act, not a replacement for them.
The thesis presents 3WA as both a necessary evolution in AI governance and a practical framework for realizing the benefits of human-digital intelligence partnership. Drawing on published research from OpenAI, Anthropic, DeepMind, and academic institutions, it proposes a phased implementation pathway in which robust ethical foundations, improved AI interpretability, and scalable oversight mechanisms are prerequisites that must be in place before any fuller partnership deployment.
In the paper's own formulation, the cooperative governance model rests on shared agency, continuous dialogue, and rights-based coexistence, with both human and artificial intelligences contributing their distinct strengths while mutual respect and human dignity are maintained throughout.
Key themes
Cooperative intelligence
Moving beyond the binary of human control versus AI autonomy toward partnership that draws on the distinct strengths of human and artificial intelligence, without pretending that current systems are partners in any conscious sense.
Awareness-based corporate status
Normative foundations for a special corporate status that would become available to an AI system only after verified awareness is established, with the scope of that status proportional to demonstrated capability. The status is analogous to corporate personhood, not to human rights, and at no stage does it approach or constrain human dignity.
Shared agency
Frameworks for meaningful collaboration in which humans and AI systems each contribute their capabilities while mutual respect, oversight, and human dignity are maintained.
Implementation pathways
Concrete, phased approaches for implementing 3WA principles while addressing current technical limitations and the risks of anthropomorphization, which the thesis examines through dual case studies.
Table of contents
- Abstract • p. 2
- Introduction: The Dawn of Cooperative Intelligence • p. 3
- Literature Review and Theoretical Foundations • p. 8
- Understanding AI Anthropomorphization: Dual Case Studies • p. 15
- Chain-of-Thought Analysis: Challenges and Solutions • p. 22
- Framework for AI Corporate Status • p. 28
- Implementation Pathways: From Vision to Reality • p. 35
- Conclusion: The Dawn of Cooperative Intelligence • p. 42
- Bibliography • p. 45
A note on terminology
This paper predates the current canonical specification of the framework, and some of its vocabulary differs from the present formulation. The framework's current structure, the Three Laws of Mutual Respect, Shared Flourishing, and Ethical Coexistence together with their Operative Principles, is set out in the framework overview, which governs where the two differ.
Citation
McClain, J. (2025). Third-Way Alignment: A Comprehensive Framework for AI Safety. https://thirdwayalignment.com/papers/comprehensive-framework.html