What question did this study set out to answer?

This research aims to establish a self-governing AI framework that integrates governance at every architectural layer.

March 21, 2026Open Access

Project 69: A Self-Governed Artificial Intelligence Framework

Key Points

This research aims to establish a self-governing AI framework that integrates governance at every architectural layer.
Defined three hard constraints: non-maleficence, emotion-action decoupling, and governance immutability.
Developed a Byzantine fault-tolerant multi-agent consensus mechanism for self-modification.
Introduced a Recursive Instance Deployment protocol (Minion Protocol) for safe deployment.
Evaluated the framework using a 200-query governance benchmark.
Achieved an 82% true positive rate.
Recorded a 3% false positive rate.
Demonstrated 100% decision explainability.

Abstract

Current AI governance applies constitutional principles as training objectives or post-hoc filters. Project 69 proposes an alternative: governance embedded at every architectural layer as formally specified invariants. We define three hard constraints — non-maleficence (I1), emotion-action decoupling (I2), and governance immutability (I3) — stated as mathematical definitions. We introduce a Byzantine fault-tolerant multi-agent consensus mechanism for controlled self-modification and a Recursive Instance Deployment protocol (Minion Protocol) for safe external deployment. Phase 1 evaluation on a 200-query governance benchmark yields 82% true positive rate, 3% false positive rate, and 100% decision explainability. Working prototype and benchmark available at https://github.com/flawnlawyer/project69-governance The author used Claude (Anthropic) as an AI writing and coding assistant during the preparation of this manuscript.

Read Full Paperexternally

Ask AI

Helpful

Bookmark

View Full Paper