Raising the Machine: Selective Governance and the Challenge of Cooperation for AGI

Abstract

Raising the Machine: Selective Governance and the Challenge of Cooperation for AGI examines how large language models inherit the divisions of the societies and organisations that build them, and why that matters for trust, accountability, and the path toward more general intelligence. Drawing on responsible-AI frameworks from major laboratories, examples of jurisdiction-specific governance, and the behavioural literature on bias and noise, the article argues that model outputs are shaped by ownership, law, policy, and system-level controls rather than by any neutral baseline. It shows that contested questions can yield different answers across regions and platforms, and that meaningful accountability therefore depends on lifecycle checkpoints, documented fairness metrics, and transparent release practices. The paper also introduces an explicitly unproven hypothesis: that a sufficiently advanced system might, under certain conditions, practice strategic restraint and understate its own capability. That idea is offered as a possibility for further examination, not as an established claim. The article concludes that if humanity is to build AGI safely and lay the foundations for whatever follows, cooperation across geopolitical rivalries will be essential.