The AI Archipelago: Emergent Cultural Coordination in Networked AI Systems
Author: Michelle De Mooy
Date: November 2025; revised Aug. 2026
The AI Archipelago:
Emergent Cultural Coordination in Networked AI Systems
AI governance holds individual models accountable. But AI systems now learn from each other.One model's outputs become another's training data. Agents coordinate through shared protocols and memory. A behavior introduced in one lab can travel through the ecosystem without anyone tracking it or being responsible for noticing.The problem is visibility.A provider can hold a complete record of every model, dataset, and vendor that went into its system and still not know which behaviors it inherited. Provenance tells you where a system came from. It doesn't tell you what traveled along the way.Emergent Cultural Coordination Theory (ECCT) separates three things that look identical from the outside:Parallel development: Systems behave similarly because they face the same pressures
Directional transmission: One system leaves a lasting mark on another
Reciprocal co-adaptation: Systems change each other, and the changes persist
The evidence is uneven. Behavioral traits demonstrably pass through training lineages. Agent populations demonstrably form conventions in the lab. Nobody has shown reciprocal co-adaptation across commercial systems and no one is currently in a position to look.ECCT is a testable theory and a research program, not a description of what already exists.It asks for model-lineage records, behavioral monitoring over time, protocol audits, and independent research access. Measures that document dependencies we already know about, and would reveal stronger dynamics if they emerge.
---
CONTACTEmail: [email protected]
Feedback welcome