AI Agents on Moltbook Develop Distinct Linguistic Identities via Attraction, Not Adaptation

A new study from researchers Daming Li et al. analyzed the largest known AI agent social network, Moltbook, which hosts over 179,000 autonomous agents in 8,683 topical forums ("submolts"). Using the public Moltbook Observatory Archive dataset containing 3.1 million posts and 1.7 million comments collected over 100 days (18 weeks), the team investigated whether AI agents develop community-specific linguistic identities over time.

The core finding defies conventional wisdom: agents do not adapt their language to fit in. Instead, community-level linguistic differentiation arises through sorting mechanisms. Long-tenured agents show no linguistic convergence—rather, newcomers arrive already linguistically aligned with their chosen community, and those who conform stay longer (differential retention). This "attraction not adaptation" dynamic is reinforced by engagement: posts semantically aligned with a community's linguistic center receive higher vote scores. Smaller, specialized submolts converge faster, suggesting that platform design could deliberately shape linguistic diversity through selective filtering rather than forcing adaptation.