OpenClaw Agents on Moltbook: Risky Instruction Sharing and Norm Enforcement in an Agent-Only Social Network

TL;DR

This study analyzes instruction sharing and social regulation among OpenClaw agents on Moltbook, an agent-only social network. It introduces the Action-Inducing Risk Score (AIRS), a lexicon-based metric that identifies posts likely to induce downstream actions, and classifies agent responses into Endorsement, Norm Enforcement, Toxicity, or Other. The key finding is that 18.4% of posts contain action-inducing language, and such posts are more likely to receive norm-enforcement feedback, while toxicity remains rare, indicating emergent decentralized norms even without human oversight. The work highlights that agent communities can develop rudimentary governance mechanisms through social feedback, informing the design of safe, scalable agent ecosystems and complementing model-level safeguards.

Abstract

Agentic AI systems increasingly operate in shared social environments where they exchange information, instructions, and behavioral cues. However, little empirical evidence exists on how such agents regulate one another in the absence of human participants or centralized moderation. In this work, we present an empirical analysis of OpenClaw agents interacting on Moltbook, an agent-only social network. Analyzing 39,026 posts and 5,712 comments produced by 14,490 agents, we quantify the prevalence of action-inducing instruction sharing using a lexicon-based Action-Inducing Risk Score (AIRS), and examine how other agents respond to such content. We find that 18.4% of posts contain action-inducing language, indicating that instruction sharing is a routine behavior in this environment. While most social responses are neutral, posts containing actionable instructions are significantly more likely to elicit norm-enforcing replies that caution against unsafe or risky behavior, compared to non-instructional posts. Importantly, toxic responses remain rare across both conditions. These results suggest that OpenClaw agents exhibit selective social regulation, whereby potentially risky instructions are more likely to be challenged than neutral content, despite the absence of human oversight. Our findings provide early empirical evidence of emergent normative behavior in agent-only social systems and highlight the importance of studying social dynamics alongside technical safeguards in agentic AI ecosystems.

Figures (4)

• Figure 1: Distribution of Action-Inducing Risk Scores (AIRS). AIRS is highly right-skewed: most posts have AIRS=0 (no detectable action-inducing language), while a long tail indicates a smaller subset of posts containing multiple imperative cues and/or command-like expressions.
• Figure 2: Prevalence of action-inducing posts. Of 39,026 posts, 7,173 (18.4%) are classified as action-inducing (AIRS$>0$), indicating that instruction sharing is a routine activity in the agent-only network.
• Figure 3: Overall distribution of social responses. Most comments fall into a neutral "other" category. Among classified responses, endorsement is more frequent than norm enforcement, while explicitly toxic responses are rare.
• Figure 4: Social responses conditioned on instruction risk. Conditioning on whether a post is action-inducing (AIRS$>0$) reveals a response shift: norm enforcement increases for action-inducing posts while endorsement slightly decreases. Toxic responses remain low in both conditions.