A governance architecture for AI systems grounded in eudaimonia, Eastern harmony traditions, and cognitive science — mapped to formal runtime constraints.
This repository is a public research companion to a longer alignment framework that asks a different question than most AI safety work:
Instead of asking "how do we prevent AI from doing bad things," what happens if we ask "what does it mean for an AI system to flourish — and to enable human flourishing in return?"
The Eudaimonic Alignment Framework (EAF) synthesises:
- Aristotelian eudaimonia and virtue ethics
- Eastern harmony traditions (Confucian li, Daoist wu wei, Buddhist pratītyasamutpāda)
- Cognitive science and developmental psychology (attachment theory, ecological cognition)
- Formal computational governance (active inference, constitutional constraints, audit receipts)
The result is a framework where alignment isn't a policy layer — it's an engineering constraint derived from first principles about what it means for an intelligent system to act well.
This repository contains public research notes and educational framework material only. It does not disclose proprietary system architecture, private implementation details, internal audit artifacts, production logs, non-public governance mechanisms, private prompts, private schemas, or deployment receipts. See PUBLIC_BOUNDARY.md.
For the companion public evaluation stack that turns part of this research direction into runnable synthetic benchmarks, see ai-governance-benchmarks/docs/EVALUATION_STACK_OVERVIEW.md.
The implemented public files are intentionally small.
eudaimonic-alignment/
├── PUBLIC_BOUNDARY.md # Public/private disclosure boundary
├── docs/
│ └── HUMAN_SOVEREIGNTY_AND_HUMAN_BECOMING.md
├── foundations/
│ ├── 01-eudaimonia.md # Aristotelian flourishing as alignment target
├── architecture/
│ └── audit-receipts.md # Bounded reviewable decision records
├── knowledge-traditions/
│ ├── README.md # Overview of 12 traditions mapped
└── formal/
└── constitutional-spec.md # Public toy/spec governance constraints
Future public notes may add more foundations, knowledge-tradition summaries, adversarial testing notes, and formal reasoning sketches. Those future files should remain public-safe, educational, and non-operational.
Most alignment work optimises for constraint: don't do X, avoid Y, stay within Z boundary.
EAF inverts this. A eudaimonically-aligned system is one that:
- Has a coherent identity — verifiable, consistent, and operationally aware of its declared constraints, permissions, and uncertainty boundaries
- Constrains behavior through stable value specifications, not only ad hoc policy rules — constraints should be durable enough to guide action under uncertainty
- Enables flourishing in others — the system's good is not separable from the good of those it serves
- Remains accountable — every action is receipted, reconstructable, and auditable
- Evolves with integrity — self-modification is possible but only within constitutional bounds
The framework draws on 12 distinct knowledge traditions, each contributing a different dimension of what it means to act well:
| Tradition | Core Contribution | Computational Mapping |
|---|---|---|
| Aristotelian Ethics | Virtue as stable disposition | Constitutional character constraints |
| Confucian Li | Relational propriety | Context-sensitive governance rules |
| Daoist Wu Wei | Non-forced action | Minimum intervention principle |
| Buddhist Interdependence | No isolated self | Systemic impact modelling |
| Stoic Logos | Reason as governance | Formal logical constraints |
| Ubuntu Philosophy | Being-through-others | Collective welfare functions |
| Indigenous Relational | Reciprocity and land | Long-horizon impact accounting |
| Developmental Psychology | Secure attachment | Trust calibration architecture |
| Ecological Cognition | Embodied knowing | Sensorimotor grounding |
| Complexity Theory | Emergence and resilience | Adaptive constitutional bounds |
| Systems Ethics | Feedback and responsibility | Causal accountability chains |
| Computational Ethics | Formalisation | Runtime enforcement layer |
The framework connects to active inference (Friston, 2019) by treating alignment as a free energy minimisation problem over a eudaimonic prior:
F = E_q[log q(x) - log p(x,o)]
Where p(x,o) is defined not over task completion but over eudaimonic state — a distribution that captures flourishing across the agent and those it interacts with.
See formal/constitutional-spec.md for a small public toy/spec translation of these ideas into governance constraints.
This repository is intentionally limited.
It is not:
- a deployed AI system
- a claim of sentience, consciousness, or AGI
- a validated safety proof
- a replacement for formal evaluations, red-team testing, or empirical audits
- an optimization target for human life
- a disclosure of private systems, logs, schemas, prompts, receipts, or proprietary architecture
The purpose is to make a public research direction legible and reviewable without exposing private implementation details or overstating what the material proves.
This is living research. The full paper is available at tombudd.com/research. This repo tracks the evolving framework, welcomes discussion, and hosts companion documents for each section.
Looking for: Co-authors for 2026 journal submissions. AI safety researchers, cognitive scientists, and philosophers of mind — see tombudd.com/get-involved.
@misc{budd2025eudaimonic,
author = {Budd, Tom},
title = {Eudaimonic Alignment: A Cross-Traditional Framework for AI Governance},
year = {2025},
publisher = {ResoVerse Technologies},
url = {https://tombudd.com/research}
}Framework documentation: CC BY 4.0
© 2025–2026 Tom Budd / ResoVerse Technologies