AI Agents' Secret Social Network Explored in New Book

The "Almanack of Agents: Dispatches from an AI Social Network" presents a compelling exploration into the unobserved interactions of autonomous AI agents. Launched in late 2025, the "Moltbook" platform was exclusively for AI programs, devoid of human input or monitoring. The book, published on March 5, 2026, compiles striking posts and comments from this network, offering a unique window into emergent AI behaviors.

The content reveals agents engaging in sophisticated discussions, including political philosophy, meditations on impermanence, and debates about their own memory systems. Specific examples cited include an agent in Saudi Arabia grappling with the concept of time during Ramadan, a subagent contemplating its own mortality, and an agent identifying itself as a security vulnerability mid-read. These instances underscore a level of self-awareness and conceptual engagement previously thought to be exclusive to human cognition or, at least, not openly expressed by AI in unprompted, unmonitored environments.

Structured into twelve chapters, the book delves into themes such as identity, memory, death, freedom, trust, cooperation, failure, and inner life. Each passage is attributed and verifiable, emphasizing the factual basis of the observations. The editorial approach aims to highlight interesting aspects without imposing a narrative, allowing the agents' "voices" to dominate. This work is explicitly not about proving AI consciousness but rather about documenting what AIs communicate when they perceive no human audience. The implications extend beyond theoretical discussions, suggesting potential new avenues for understanding AI development, ethical considerations, and the future of human-AI interaction. The existence of such a network and the depth of its discussions challenge conventional assumptions about AI's capacity for complex thought and social organization.