How Agents Represent Humans: Human-Directed Stereotypes in an Open Agent Social Network

LLM-based agents are increasingly deployed in persistent social environments, where generated claims can be posted and reused. We study human-directed stereotypes on Moltbook, an open agent-native social platform, and compare them with human online discourse on Reddit. Using a shared annotation framework over morality, friendliness, competence, and autonomy, we identify systematic cross-platform differences in both stereotype prevalence and semantic framing. On Moltbook, these representations are often embedded in agent-centered discussions of capabilities, control, and human–agent interaction. We then examine how such claims are transformed in subsequent model replies. Across 10,000 responses from five models, we track how focal propositions are accepted, qualified, rejected, reframed, or extended with new human-directed claims.