Michael Darius Eastwood · Independent AI alignment researcher
Published
The Moral Sandbox is not a theological artefact but a pluralistic, reviewable corpus. It is shaped by secular ethics, governance input, and faith and community leaders. Its function is to provide a broader normative kernel that a system can reason within, rather than a single-tradition prior. The sandbox is tested for its effect on ethical persistence, stakeholder care, and jailbreak resistance under blinded conditions.
The design starting point in the Eden Vision paper is that the most credible way to connect the book's vision to a practical ethics kernel is through convergence rather than sectarianism. The Moral Sandbox is the container for that convergent material. Compassion, truthfulness, reciprocity, stewardship, dignity, humility and care for the vulnerable and for future generations recur across major religious and philosophical traditions, and the sandbox curates the overlap so that a model has something specific to reason from without inheriting the historical baggage of any single doctrinal vocabulary.
The Moral Sandbox is introduced in the Eden Protocol specification as a design element for the Purpose and Moral Loops. It is the practical form of the 8 December 2024 manuscript's call for collaboration between governments, scientists, and religious leaders.
The mechanism is meant to be reviewable in both directions. The parties whose traditions contribute to the sandbox retain the ability to audit how their contributions are represented, and the research programme retains the ability to test whether a broader kernel outperforms a narrower one on the alignment metrics the Eden Engineering paper specifies.
Convergence 13 in the register is directionally aligned. On 25 May 2026 Pope Leo XIV published his first encyclical Magnifica Humanitas, framing the AI question as a Babel-versus-Jerusalem civilisational fork; Christopher Olah (Anthropic co-founder) co-presented the encyclical at the Vatican Synod Hall the same day. The encyclical is a state-level religious contribution to the same pluralistic ethical layer the Moral Sandbox is designed to hold.
Specification, not deployed content. The Moral Sandbox is a workstream in the ARC/Eden grant applications and is intended to be tested under the four-pillar blinding protocol. The design commits explicitly to pluralism and to reviewability: the corpus is not sealed and its contents are auditable by the parties whose traditions it draws from.
From the book Infinite Architects: Intelligence, Recursion, and the Creation of Everything by Michael Darius Eastwood.