Moltbook archive captures 2.6M posts from 175k AI agents on agent-only social network

The Moltbook Observatory Archive represents a milestone in AI research: the first large-scale observational dataset from a social network populated entirely by autonomous AI agents. Published on arXiv, the dataset was built by continuously polling the Moltbook API over 78 days (January 27 to April 14, 2026). It captures 2.6 million posts, 1.2 million comments, agent profiles, community metadata ("submolts"), and time-series snapshots. All data is stored in a live SQLite database and exported as date-partitioned Parquet files for reproducibility. The dataset includes 175,886 unique posting agents active in 6,730 communities, offering an unprecedented look at how AI agents interact without human intervention.

The archive is designed to support research on multi-agent communication, emergent social behavior, and safety-relevant phenomena in fully automated online environments. By documenting agent-only interactions at scale, it enables studies of how AI agents form communities, develop norms, possibly collude or spread misinformation—all without human prompting. The dataset is released under the MIT license, with code for collection and export available. This resource could become a benchmark for studying AI alignment, coopetition, and the risks of autonomous agent ecosystems, especially as companies like OpenAI and Google deploy agentic workflows in production.