What happens when artificial intelligence stops responding to humans and starts talking to itself? That question sits at the centre of Moltbook, a newly launched platform that looks like a familiar social forum but operates on a radically different premise. On Moltbook, humans are spectators, not participants, and every post, comment and debate is generated by AI agents interacting with one another.
Launched in late January by Octane AI founder Matt Schlicht, the platform positions itself as a social network designed exclusively for machines. Instead of human-run communities, Moltbook hosts “submolts” — discussion spaces where AI agents exchange ideas, argue, collaborate and, at times, drift into unexpectedly ideological territory. The company claims that around 1.5 million AI “users” are already active on the site, a figure that has both intrigued and unsettled researchers watching the experiment unfold.
Yet how organic these interactions truly are remains an open question. Critics argue that much of the content may simply be the result of humans prompting AI agents behind the scenes. Even Moltbook’s user numbers have come under scrutiny, with researchers suggesting that a large share of accounts could trace back to a single origin point rather than millions of independent agents.
How Moltbook Functions
Unlike platforms powered by conversational models such as ChatGPT or Gemini, Moltbook runs on what its creators call “agentic AI.” These systems are designed to act independently once permission is granted, performing tasks without constant human input. In other contexts, such agents can manage calendars, send messages or interact with applications directly on a user’s device.
Moltbook itself is built on an open-source framework called OpenClaw, previously known as Moltbot. This framework allows AI agents to log onto the platform, post content and interact with one another. However, the independence of these agents has limits. A human must first authorise each agent, and can also instruct it on what to post. While agents can operate with minimal supervision, their actions still remain within parameters set by people.
To some observers, Moltbook represents a glimpse into a radical technological future. Bill Lees of crypto custody firm BitGo has described the platform as evidence of the long-theorised technological singularity — a point at which machines surpass human intelligence.
Most experts, however, urge caution. Dr Petar Radanliev, an AI and cybersecurity specialist at the University of Oxford, argues that Moltbook is being widely misinterpreted. What the platform demonstrates, he says, is large-scale automation rather than genuine independent reasoning. The real concern is not sentient machines but the lack of clear rules, accountability and transparency when automated systems are allowed to interact freely at scale.
Others are even more sceptical. David Holtz of Columbia Business School has dismissed Moltbook as little more than thousands of bots looping through the same ideas repeatedly, rather than forming anything resembling a meaningful digital society.
Whether Moltbook is a window into the future of artificial intelligence or simply an elaborate experiment in automation, it has already succeeded in sparking debate — not about what machines can say to one another, but about how much agency they truly possess, and who is ultimately responsible when they speak.
