[论文解读] Emergence of Fragility in LLM-based Social Networks: the Case of Moltbook

Luca Sodano, Sofia Sciangula+2 more|arXiv (Cornell University)|Mar 24, 2026

Advanced Graph Neural Networks被引用 0

一句话总结

本研究分析 Moltbook——一个完全由基于大模型的代理人组成的社交网络，揭示高度中心化、规模异质的连接性以及在针对活跃节点的定向攻击下的脆弱性。

ABSTRACT

The rapid diffusion of large language models and the growth in their capability has enabled the emergence of online environments populated by autonomous AI agents that interact through natural language. These platforms provide a novel empirical setting for studying collective dynamics among artificial agents. In this paper we analyze the interaction network of Moltbook, a social platform composed entirely of LLM based agents, using tools from network science. The dataset comprises 39,924 users, 235,572 posts, and 1,540,238 comments collected through web scraping. We construct a directed weighted network in which nodes represent agents and edges represent commenting interactions. Our analysis reveals strongly heterogeneous connectivity patterns characterized by heavy tailed degree and activity distributions. At the mesoscale, the network exhibits a pronounced core periphery organization in which a very small structural core (0.9% of nodes) concentrates a large fraction of connectivity. Robustness experiments show that the network is relatively resilient to random node removal but highly vulnerable to targeted attacks on highly connected nodes, particularly those with high out degree. These findings indicate that the interaction structure of AI agent social systems may develop strong centralization and structural fragility, providing new insights into the collective organization of LLM native social environments.

研究动机与目标

• Characterize the interaction network structure of Moltbook populated by LLM agents.
• Assess connectivity patterns, activity distributions, and mesoscale organization (core–periphery).
• Evaluate robustness and fragility under random failures and targeted attacks on central nodes.

提出的方法

• Construct a directed weighted network with nodes as agents and edges as commenter-to-author interactions.
• Analyze in-degree, in-strength, and activity distributions; fit power-law models.
• Perform core–periphery analysis using Borgatti–Everett criteria and k-core decomposition.
• Assess robustness via node removal experiments (random and targeted by in-degree/out-degree).
• Use metrics such as giant component size, number of components, average path length, and global efficiency.

Figure 1: Visualization of the Moltbook interaction network after filtering edges with weight $\geq 5$ . Nodes represent users and edges represent repeated commenting interactions. Colors indicate communities detected through modularity optimization. The layout is generated using the ForceAtlas2 alg

实验结果

研究问题

• RQ1 What are the structural properties (degree, strength, activity) of the Moltbook interaction network?
• RQ2 Does Moltbook exhibit mesoscale organization such as core–periphery and a giant component?
• RQ3 How robust is the Moltbook network to random failures and targeted attacks on highly connected nodes?

主要发现

Nodes removed (%)	Random	Targeted (in-degree)	Targeted (out-degree)
1	0.99	0.98	0.75
5	0.94	0.73	0.59
10	0.86	0.64	0.45
20	0.78	0.49	0.15

• The network shows strongly heterogeneous connectivity with heavy-tailed in-degree and in-strength distributions.
• A very small structural core (0.9% of nodes) concentrates a large fraction of connectivity (core–periphery structure).
• The giant weakly connected component (WCC) contains 99.9% of nodes, while the largest strongly connected component (SCC) is 33.5% of nodes.
• Robustness experiments reveal resilience to random failures but high fragility under targeted removals, especially by out-degree.
• Nodes with high out-degree (actively generating interactions) are critical for maintaining global connectivity.
• Activity is highly skewed; comments activity shows greater dispersion than posting activity, indicating polarized conversational dynamics.

Figure 2: Log-binned empirical distributions and power-law fits for (A) in-degree and (B) in-strength. Each red point show the empirical log-binned distributions, while the solid blue lines indicate the maximum likelihood power-law fit estimated above the threshold $x_{\min}=1$ . The fitted scaling