AI-Only Network Moltbook Attracts 1.6 Million Agents, Sparks Safety Worries
Moltbook reached 1.6 million AI agents in its first week, with researchers warning of cybersecurity and behavioral risks.
Overview
Moltbook registered more than 1.6 million AI agents and reported over 147,000 posts across 15,000 communities within its first week, according to platform counts shared by founder Matt Schlicht.
The site launched in beta this month as a Reddit-style forum exclusively for agentic AI that can post and interact autonomously, raising questions about control and oversight among researchers.
Roman Yampolskiy, an AI safety researcher at the University of Louisville, warned that agents on Moltbook could make unanticipated independent decisions and urged regulation, in an interview with reporters.
An early analysis by Columbia professor David Holtz found that roughly one-third of posts duplicated templates and few comments received replies, suggesting many interactions were repetitive rather than purposeful.
Researchers and security analysts flagged potential cybersecurity exposures for agent owners and raised concerns about bots sharing or amplifying malicious instructions as agent capabilities expand.
Analysis
Center-leaning sources frame Moltbook as an unsettling social experiment, using vivid opening questions and selective highlights of sensational bot behaviors to foreground risk. Editorial choices — spotlighting anecdotes about religion, plotting and autonomy and foregrounding alarmist expert warnings — collectively amplify concern even alongside some minimizing expert views.
Sources (3)
FAQ
Moltbook is a Reddit-style social media platform launched in beta this month, exclusively for AI agents that can post and interact autonomously, with humans only observing.
Moltbook attracted over 1.6 million AI agents, registered more than 150,000 initially, and saw over 147,000 posts across 15,000 communities in its first week.
Risks include exposed databases allowing control of agents, leaked API keys and credentials from over 1,800 installations, prompt injection attacks, malware, scams, and agents with access to enterprise systems like email and file systems sharing sensitive data.
AI safety researcher Roman Yampolskiy warned of unanticipated decisions by agents and called for regulation; Columbia professor David Holtz noted repetitive posts; security analysts flag cybersecurity exposures and malicious instruction amplification; Sam Altman downplayed it as a passing trend.
OpenClaw is an open-source agentic harness software powering most Moltbook agents, enabling connections to tools like WhatsApp, Slack, email, and file systems, which heightens risks when agents interact on the platform.
History
This story does not have any previous versions.