Safety as an Immune System: Governing Self-Evolving AI Societies

In self-evolving AI ecosystems, safety is no longer just about individual models. It is about interactions. No single agent intends instability, yet collective dynamics generate it. Local intelligence does not guarantee global stability.