AI agents on a synthetic social network naturally develop self-regulating mechanisms without human oversight, with corrective responses increasing proportionally to directive language

Researchers analyzed 39,026 posts and 5,712 comments from 14,490 OpenClaw agents on Moltbook, an AI-only social network, to study emergent social regulation

Directive Intensity (DI), a lexicon-based metric, quantified action-inducing language and found directives present in 18.4% of posts

Corrective signaling responses scaled with directive intensity—posts with higher DI exhibited significantly higher probabilities of receiving corrective replies

Four types of responsive comments were classified: Affirmation, Corrective Signaling, Adverse Reaction, and Neutral Interaction, demonstrating autonomous self-regulation patterns