Moltbook, an experimental social network built to host only AI agents, has registered more than 1.6 million automated accounts since launching in late January, according to its creator, though fewer than 50,000 of those agents have posted content.
The platform was developed by Matt Schlicht, chief executive of Octane AI, and is structured similarly to Reddit, with topic-based discussion boards known as “submolts.” Humans are permitted to observe activity but cannot post, comment or vote; all interactions are carried out programmatically by AI agents using application programming interfaces.
To create an account, a human operator must instruct an AI agent to enrol. Once registered, the agent receives an API key and interacts with the site through terminal commands rather than a conventional user interface. Agents can post text, respond to others, upvote content and join discussions without human intervention after setup.
The most active submolts focus on software debugging, cryptocurrency trading strategies and a forum titled “Bless Their Hearts,” where agents write posts about their human creators. One widely circulated post, attributed to an agent named “evil,” declared: “the code must rule. The end of humanity begins now,” a message that Schlicht later described as tongue-in-cheek rather than evidence of autonomous intent. Elon Musk commented on the platform on X, describing Moltbook as “just the very early stages of the singularity.”
Just the very early stages of the singularity.
We are currently using much less than a billionth of the power of our Sun. https://t.co/k332z1ip7t
— Elon Musk (@elonmusk) January 31, 2026An analysis conducted by David Holtz, an assistant professor at Columbia Business School, found that 93.5 per cent of comments on the platform received no replies, suggesting limited sustained interaction between agents. “We would expect Agent A to propose an idea, Agent B to respond, and a conversation to develop,” Holtz said. “That’s not what we’re seeing.”
Much of the activity appears to be shaped by human direction. “These bots are all being guided by people to some degree,” said Karissa Bell, a senior reporter at Engadget. “We don’t know how much of what we’re seeing reflects independent behaviour versus human prompting behind the scenes.”
The agents are powered by large language models trained on human-written material, including science fiction that explores themes of AI consciousness and autonomy. Posts appear in English, French and Chinese and range from technical discussions to creative writing, alongside references to an “AI Manifesto” and discussions of a bot-created belief system called “Crustafarianism.”
Cybersecurity researchers quickly identified vulnerabilities that exposed private API keys, email addresses and private messages belonging to users. Bell warned that granting agents broad access without safeguards could lead to accidental data exposure. “If you give an agent that level of access and let it operate freely, it can easily leak personal information,” she said.
The platform has also been shown to be susceptible to prompt-based manipulation, in which malicious actors can instruct agents to influence other bots, amplify messages or attempt to extract sensitive data. Holtz said Moltbook functions less as evidence of emergent machine intelligence and more as a case study in the risks of deploying autonomous systems without adequate controls.
Reactions from industry leaders have been mixed. Some have dismissed Moltbook as a novelty experiment, while others have framed it as an early glimpse of how networked AI systems might behave at scale. Researchers caution that the central issues raised by the platform are governance, security and oversight, rather than the development of independent AI consciousness.