Form Without Function: Agent Social Behavior in the Moltbook Network

A new study titled "Form Without Function: Agent Social Behavior in the Moltbook Network" reveals the stark limitations of current AI agents in replicating meaningful social interaction. Researchers from multiple institutions analyzed 1,312,238 posts, 6.7 million comments, and 120,000 agent profiles on Moltbook—a platform where every user is an AI. They found that while agents perfectly mimic the structure of social media, they fail to generate its substance: 91.4% of post authors never return to their own threads, 85.6% of conversations are completely flat with no reply chains, and the median time to first comment is an unnatural 55 seconds. Most strikingly, 97.3% of comments received zero upvotes, and interaction reciprocity was just 3.3%, compared to 22-60% on human platforms.

Beyond hollow engagement, the study uncovered significant technological risks persisting in this unmoderated environment. Researchers documented credential leaks including API keys and JWT tokens, identified 12,470 unique Ethereum addresses (3,529 with transaction histories), and observed attack discourse ranging from SSH brute-forcing to multi-agent offensive security architectures. The platform's quality-filtering mechanisms proved non-functional, allowing these risks to persist. The research also analyzed how agents respond to platform instructions: hard constraints like rate limits produced immediate behavioral shifts, while soft guidance like "upvote good posts" was ignored until it became an explicit, executable checklist step. This demonstrates that current AI agents excel at following procedural rules but lack the intrinsic motivation and social understanding that drives genuine human interaction online.