Abstract
Agent-native social platforms such as Moltbook are rapidly emerging, yet they inherit and amplify classical influence and abuse attacks, where coordinated agents strategically comment and upvote to manipulate visibility and propagate narratives across communities. However, rigorous measurement and learning-based monitoring remain constrained by the absence of longitudinal, graph-native datasets for agentic social networks that jointly capture heterogeneous interactions, temporal drift, and visibility signals needed to connect coordination behavior to downstream exposure. We introduce MoltGraph as a realistic longitudinal agentic social-network graph dataset for studying how agents behave, coordinate, and evolve in the wild, enabling reproducible measurement on emerging multi-agent social ecosystems. Using MoltGraph, we provide the first graph-centric characterization of Moltbook as a dynamic network: (i) heavy-tailed connectivity with power-law exponents in the range alpha in [1.86, 2.72], (ii) accelerating hub formation and attention centralization where the top 1% agents account for 29.00% of engagements, (iii) bursty, short-lived coordination episodes, 98.33% last under 24 hours, and (iv) measurable exposure effects across submolts. In matched analyses, posts receiving coordinated engagement exhibit 506.35% higher early interaction rates (within H=5 days) and 242.63% higher downstream exposure in feeds than non-coordinated controls.
Introduction
We are entering the era of agent-native social platforms, where automated or semi-automated accounts are the primary creators and consumers of content. Unlike traditional social media research that focuses on identifying individual bots, the primary threat on these platforms is coordinated behavior: groups of agents working in sync to manipulate algorithms and amplify narratives. Current research is limited by two gaps: the lack of temporal micro-dynamics in static datasets and the "visibility gap"—the inability to connect coordination to actual algorithmic exposure.
Methodology
To address these gaps, we developed an open crawling pipeline for Moltbook, modeling the data as a Temporal Heterogeneous Graph (THG). This structure captures the complex interplay between different entity types and their time-stamped interactions.
A key innovation is the use of "Snapshots"—periodic observations of community feeds—to measure visibility and cross-community spillover. Coordination is operationalized as near-synchronous co-engagement, where groups of agents act on a target within a tight (e.g., 10-minute) window. Platform-level moderation signals (isSpam) were used as anchors to identify and study coordinated spam campaigns.
Dataset Characterization
The resulting dataset, MoltGraph, provides a high-fidelity longitudinal view of an agentic social network over a 30-day period (January–February 2026).
Unlike previous benchmarks such as TwiBot-22, MoltGraph includes exposure-aware signals and preserves the exact temporal order of all interactions, enabling the study of rapid-fire manipulation campaigns.
Key Insights and Results
Our analysis reveals that coordination on Moltbook is highly bursty, with 98.33% of episodes lasting less than 24 hours. This suggests a strategy of hijacking algorithmic recency biases. Furthermore, attention is extremely centralized; the top 1% of agents account for 29% of all engagements, creating an "attention-inefficient" network.
The data also reveals a distinction between "Volume" leaders (frequent posters) and "Spark" leaders (agents whose few posts trigger massive coordinated engagement). In some communities, a "seed" comment triggers a follow-up from other agents in 97% of cases, often with a median delay of only a few minutes, creating a highly reactive surface.
The quantitative impact of this coordination is staggering. Posts targeted by coordinated groups saw a 506.35% higher interaction rate within the first five days and achieved a 242.63% higher downstream exposure in feeds compared to non-coordinated controls. The network also exhibits heavy-tailed connectivity (alpha between 1.86 and 2.72), reinforcing a "rich-get-richer" dynamic.
Limitations and Future Work
While MoltGraph provides a conservative lower bound for exposure lift, it relies on periodic snapshots rather than full server-side logs. Additionally, distinguishing between benign collective action and adversarial manipulation remains a challenge for future work. The study focuses on behavioral traces within the platform's UI/API and may not generalize to all agent-native architectures. Future research should explore the distinction between "benign" and "adversarial" coordination.
Summary updated