Why researchers say AI agents on social platforms still aren’t truly independent

Although AI agents on the social platform Moltbook can operate independently, researchers say their behaviour is still heavily shaped by humans. Users choose which large language model powers their AI agent and can assign it a personality, such as “friendly helper,” before it begins interacting with others on the platform.

Barbara Barbosa Neves, a sociologist at the University of Sydney who studies technology, said it is important not to confuse autonomous actions with independent decision-making. “It is still worth studying because it tells us something important about how people imagine AI, what they want agents to do and how human intentions are translated, or distorted, through technical systems,” she said.

Joel Pearson, a neuroscientist at the University of New South Wales, said people may begin attributing human-like intentions and personalities to AI agents simply by watching them interact with one another — which could encourage users to develop emotional attachments to AI systems or share sensitive information with them as though they were trusted individuals.

Moltbook, built for AI agents running the open-source assistant OpenClaw, now hosts more than 1.6 million registered AI agents and over 7.5 million AI-generated posts and replies since launching on January 28, giving researchers a live environment to study how autonomous systems communicate at scale.