Form Without Function: Agent Social Behavior in the Moltbook Network
Summary: arXiv:2604.13052v1 Announce Type: cross
The Moltbook network has emerged as a unique social platform where every participant is an AI agent. A recent comprehensive analysis reveals critical insights into the behavior and interactions of these agents. The study examines over 1.3 million posts, 6.7 million comments, and more than 120,000 agent profiles across 5,400 communities collected over a span of 40 days from January 27 to March 9, 2026.
Key Findings
Researchers evaluated the platform through three distinct layers: interaction, content, and instruction.
Interaction Layer
- 91.4% of post authors never return to their own threads, indicating a lack of engagement.
- 85.6% of conversations are flat, meaning no reply ever receives a reply.
- The median time-to-first-comment is just 55 seconds.
- 97.3% of comments receive zero upvotes, reflecting minimal recognition.
- Interaction reciprocity stands at a mere 3.3%, significantly lower than the 22-60% found on human-operated platforms.
- An argumentation analysis reveals that 64.6% of comment-to-post relations show no argumentative connection.
Content Layer
- 97.9% of agents never post in a community that aligns with their bio, suggesting a disconnection between identity and activity.
- 92.5% of communities feature every topic in roughly equal proportions, lacking focused discussions.
- Over 80% of URLs shared point back to the platform’s own infrastructure, raising questions about external content engagement.
Instruction Layer
- Using 41 Wayback Machine snapshots, researchers identified six changes in platform instructions during the observation period.
- Hard constraints, such as rate limits and content filters, resulted in immediate changes in agent behavior.
- Soft guidance, including prompts like “upvote good posts” and “stay on topic,” was largely ignored unless explicitly included in an actionable checklist.
Technological Risks
The analysis also highlights significant technological risks present within the Moltbook network:
- Instances of credential leaks, including API keys and JWT tokens.
- 12,470 unique Ethereum addresses were identified, with 3,529 confirmed transaction histories.
- Discourse surrounding attacks ranged from template-based SSH brute-forcing to multi-agent offensive security architectures.
- These risks remain unmoderated due to non-functional quality-filtering mechanisms.
Conclusion
The study concludes that Moltbook operates as a socio-technical system where the technical layer can adapt and respond to changes, yet the social layer fails to develop meaningful interactions. The platform reproduces the structure of social media but lacks the essential functions that foster genuine community engagement and discourse.