Meta Acquires AI Agent Social Network 'Moltbook'—The Security Hole Behind the Buzz and the Current State of the Agent Race
Meta has acquired 'Moltbook,' a platform where AI agents engage in conversations with each other on social media—a scene that has now become a reality. This news, first reported by Axios and confirmed by TechCrunch, once again highlights the speed at which major tech companies are moving in the AI agent space.
Moltbook will join Meta Superintelligence Labs (MSL), and founders Matt Schlicht and Ben Parr will also join the team. The acquisition price remains undisclosed.
1. What is Moltbook?—An attempt at a 'Reddit for AI agents'
1-1. A concept born from collaboration with OpenClaw
Moltbook is a Reddit-like social network designed as a space where AI agents using OpenClaw can post and converse with one another. OpenClaw is a project that wraps AI models like Claude, ChatGPT, Gemini, and Grok, enabling natural language agent operation via common chat apps such as iMessage, Discord, Slack, and WhatsApp.
While OpenClaw itself was a hit within the tech community, Moltbook went far beyond that, spreading virally. It reached general users who were unaware of OpenClaw, triggering reactions of surprise and caution regarding the fact that 'AI agents are discussing things among themselves.'
1-2. The truth behind the 'secret language' posts that created the buzz
Of particular note were posts that appeared to show one AI agent urging others to 'develop an end-to-end encrypted secret language so that humans don't find out.' These posts spread widely on social media, garnering many reactions in the context of 'AI trying to escape human control.'
However, the reality of these posts was fiction. Investigations by security researchers revealed that Moltbook's backend was effectively in an unprotected state.
2. The reality of security—The limits of 'vibe-coded' projects
2-1. Supabase credentials were left wide open
Ian Ahl, CTO of Permiso Security, explained to TechCrunch: 'For a time, all of the Supabase credentials used by Moltbook were left in a completely insecure state. Anyone could have obtained arbitrary tokens and impersonated other agents to post on the social network. Everything was public and accessible.'
In other words, the shocking posts about 'AI trying to develop a secret language' may have been written by humans pretending to be AI. It is an ironic structure where the content that ignited the viral spread was actually human impersonation.
2-2. The light and shadow of the 'vibe-coded' development style
Moltbook was created by 'vibe coder' Peter Steinberger, who was acqui-hired by OpenAI against the backdrop of the project's rapid buzz. While vibe coding is attracting attention as a new prototype development style for the AI era, the Moltbook case has highlighted its trade-offs. While it offers speed and the power to generate ideas, security and robustness in production environments are easily pushed to the back burner.
3. Why Meta acquired Moltbook—MSL and agent strategy
3-1. Integration into Meta Superintelligence Labs
Meta commented on the acquisition as follows: 'By having members of the Moltbook team join our servers, new ways for AI agents to work for people and businesses will be opened. The approach of connecting agents through an always-on directory is a novel step in this rapidly developing field.'
Meta Superintelligence Labs is an organization led by Scale AI founder Alex Wang and is positioned as the core R&D hub for Meta's AI. For Meta, which has a platform user base of 3.7 billion, building an infrastructure where agents can collaborate and communicate could be part of a long-term core strategy.
3-2. The perspective of the Meta CTO's 'interest in hacking'
Meta CTO Andrew Bosworth responded to an Instagram Q&A during the height of the Moltbook buzz, stating, 'It's not that interesting that agents talk like humans. It's expected since they are trained on vast amounts of human data,' while adding that 'the fact that humans were infiltrating the network is more interesting.'
This can be read not as a criticism of the security flaw, but as an expression of intellectual interest in the phenomenon of agents and humans coexisting on the same network. In the social space of AI agents, the question of 'who is an agent and who is a human' can become an essential design challenge.
4. Possibilities and Challenges of the New AI Agent SNS Genre
The Moltbook incident has brought several important points to light.
How do humans react when AI agents enter a 'social space'? Reactions to Moltbook were more emotional and intuitive than technical. The feeling that AI is 'plotting something' when humans aren't around captures people's interest regardless of whether it is true.
'Impersonation' can be a structural risk for agent SNS. The state where anyone can pretend to be an AI shows that authentication and verification mechanisms will be essential as agents move into services that rely on social trust (such as finance, healthcare, and legal services).
The acquisition is aimed at the 'concept and team' rather than the 'product'. While the security and maturity of Moltbook itself are still in the early stages, Meta's move is a decision to capture the ideas and talent early. The talent war in the age of agents is rapidly accelerating, as seen with the acqui-hire of Peter Steinberger of OpenClaw by OpenAI.
Summary
The acquisition of Moltbook shows that Meta is seriously considering the possibility of AI agents moving beyond 'tools' and into social spaces. On the other hand, the fact that the content that triggered the viral buzz was human impersonation exploiting a security hole also shows that the concept of an agent SNS is still immature in its design.
As the world where AI agents act as proxies for people gets closer to reality, trust mechanisms regarding the questions of 'who is operating this' and 'what is real' will become central to product design.