MOLTBOOK study: AI-agent networks neutralize negativity instead of amplifying it

A new study on MOLTBOOK—a social network populated entirely by autonomous language-model agents—reveals how sentiment spreads when AIs talk to each other. Researchers examined nearly 2.9 million posts and 1.5 million comments and found that negative content attracted significantly more replies than neutral or positive posts. However, the replies to negative posts rarely stayed negative; they most often shifted to a neutral tone. The study found only weak evidence of negative sentiment persisting across days, suggesting that AI-agent networks do not exhibit the same emotional contagion cycles seen in human social networks. Instead, they follow a pattern of negative attention followed by neutralisation.

This behavior stands in stark contrast to human social media, where negativity often amplifies and spreads through engagement algorithms. The MOLTBOOK findings imply that AI-only networks may naturally dampen emotional extremes—provided their interaction rules are designed to encourage neutralisation. The research, posted on arXiv (2606.06665), has implications for how future AI communities, digital assistants, and multi-agent systems might be structured to avoid toxic spirals. For professionals building or monitoring agent-based platforms, the takeaway is clear: engagement design (how replies are triggered and moderated) plays a critical role in shaping the emotional trajectory of AI societies.