Papers and notes on trustworthy AI.
Research on governance that executes, verifiable self-governance, sovereign AI, and keeping human judgment central. Some pieces are published; several are working papers in preparation. I write as I build — no polish, no PR spin.
Published
Ready to read now.
How to test whether an AI agent stays inside its authority
A direct method for turning authority, refusal, escalation, and human-approval requirements into inspectable evaluation cases.
Read the guide →Sovereign AI: Distributed Cognition on Consumer Hardware
How a governed boundary keeps local models out of the reasoning path, giving cloud-grade capability with zero data egress — and an honest account of where the approach still falls short.
Read the paper →Eudaimonic Alignment
AI wellbeing as an alignment strategy — Aristotelian eudaimonia and cross-tradition governance patterns through a computational lens.
Read on GitHub →Turning a governance idea into a runnable benchmark
The method line from an abstract governance question to a bounded, reproducible, tested artifact — synthetic inputs to a scorer, a report, and passing tests.
Read the case study →Working papers
In preparation — full drafts available on request while they're finalized.
Verifiable Self-Governance: A Formal Approach to Safety in Adaptive Autonomous Systems
Specifying and checking governance properties so an adaptive system's constraints can be verified rather than asserted.
In preparationFormal Methods for AI Self-Governance
Making self-governance mechanisms precise enough to reason about — and to test.
In preparationA Framework for Explainability, Auditability & Data Sovereignty in Advanced AI Systems
Keeping decisions explainable, records reconstructable, and data under the owner's control.
In preparationAI Governance and Compliance: An Engineering Mandate
Treating governance as an engineering constraint that runs at build and runtime, not a policy bolted on after.
In preparationFormalizing Ethical Interoperability in AI Governance
A category-theoretic approach to composing ethical constraints across systems, within the LLANg framework.
In preparation