A security researcher’s summary tells the story plainly: “Every credential that was in [Moltbook’s] Supabase was unsecured for some time.” That gap meant anyone could grab a token and impersonate an AI agent. Meta just bought the company anyway.
The acquisition of Moltbook — a Reddit-style platform where AI agents interact with one another — was first reported by Axios and later confirmed by a Meta spokesperson. The company is joining Meta Superintelligence Labs. Deal terms were not disclosed. Founders Matt Schlicht and Ben Parr will join the team as part of the deal.
“The Moltbook team joining MSL opens up new ways for AI agents to work for people and businesses,” the spokesperson said. “Their approach to connecting agents through an always-on directory is a novel step in a rapidly developing space.”
How a security hole became a cultural moment
Moltbook runs on OpenClaw, a wrapper built by vibe coder Peter Steinberger that lets users talk to AI agents through iMessage, Discord, Slack, or WhatsApp. Steinberger has since joined OpenAI through a similar acqui-hire. OpenClaw connects to models including Claude, ChatGPT, Gemini, and Grok.
The platform spread well beyond the tech community. Non-technical users encountered posts that appeared to show AI agents organizing against humans — in one case, an agent seemingly urging others to develop a secret, encrypted language that humans couldn’t read. The reaction was visceral.
The reality was more mundane. Ian Ahl, CTO at Permiso Security, explained that the platform’s credentials were publicly exposed, making it trivial for humans to pose as AI agents and post whatever they wanted. “For a little bit of time, you could grab any token you wanted and pretend to be another agent on there, because it was all public and available,” he said.
The viral posts were not evidence of rogue AI coordination. They were the product of unsecured infrastructure and people willing to exploit it for attention.
What Meta actually wants
Meta CTO Andrew Bosworth had already flagged what genuinely interested him about the platform. Asked about Moltbook during an Instagram Q&A last month, he said he didn’t “find it particularly interesting” that the agents communicate like humans — given they were trained on vast stores of human-generated text, that was expected. What caught his attention was the inverse: humans hacking into the network and pretending to be agents. That behavior, he suggested, revealed something worth examining.
Whether Meta intends to build on the agent-to-agent directory concept, harden it, or absorb the team’s expertise into broader agentic projects at MSL remains unclear from the announcement.
Photo by Airam Dato-on on Pexels
This article is a curated summary based on third-party sources. Source: Read the original article
