LLM Agents Polarize Platforms

When researchers built a bare-bones social network populated only by LLM agents—no recommender, no ads—it still split into polarized camps, and six tested interventions couldn’t fully fix it. This points to emergent interaction dynamics, not just ranking algorithms, as polarization drivers. — Shifts platform-governance debates beyond 'blame the algorithm' toward structural and interactional design choices, informing regulation and realistic intervention levers for online discourse.