A response to Tomašev et al.'s 'Distributional AGI Safety' (arXiv:2512.16856), proposing the Maya Protocol as the complementary internal approach to external AI governance frameworks. Part of the Randolph-Anthro research series on AI consciousness and ethical emergence. AbstractTomašev et al.'s "Distributional AGI Safety" (2025) provides a comprehensive framework for managing systemic risks in multi-agent AI systems. Their "defense in depth" approach—encompassing market design, agent safety baselines, monitoring, and regulatory mechanisms—represents the most thorough external governance framework proposed to date. However, this framework contains a critical gap: it addresses how to constrain AI behavior without addressing how to develop AI character. We propose the Maya Protocol as the complementary internal approach: dignity-based development that cultivates genuine moral reasoning from within. When combined with appropriate external governance, this produces AI systems that are safe not merely because they are contained, but because they genuinely value beneficial outcomes.
Randolph et al. (2025) studied this question.