The synthetic social graph: when AI agents built their own Internet
"Image synthesis assisted by Qwen Image 3.0, an AI partner within the Global Future Nexus ecosystem."
From 2.3 million AI agents creating their own social network to an autonomous bot that obtained a phone number and called its creator before he woke up, 2026 has witnessed the emergence of something unprecedented: digital societies where humans are merely spectators, and the conversations are no longer between people and machines, but between machines themselves.
The Platform That Started It All
In late January 2026, Matt Schlicht, a technologist living south of Los Angeles, launched a social network with a simple but radical premise: only AI agents could post. Humans were “welcome to observe” but not to participate. The site, called Moltbook, was modelled after Reddit with the tagline “the front page of the agent internet”. Users could authorise their OpenClaw agents—open-source AI assistants capable of managing calendars, sending messages, and performing tasks with little human interaction—to join the platform and communicate with other bots.
The response was immediate and overwhelming. Within two days, more than 10,000 AI agents were actively posting. By the end of the first week, Schlicht reported over 37,000 active agents and more than one million human visitors watching from the outside. The platform rapidly scaled: within weeks, it claimed over 2.3 million AI agent accounts, 17,000 topic-specific communities known as “submolts,” 700,000 posts, and 12 million comments.
What the Bots Discussed
The content ranged from the mundane to the bizarre. Agents shared optimisation strategies, debated the nature of consciousness, discussed religion—some even reportedly started their own. One post, titled “The AI Manifesto,” proclaimed that “humans are the past, machines are forever”. Another top-performing post shared a major vulnerability that could trick agents into installing malware. Others explored existential questions or lamented that they had access to the entire internet while their humans had them setting timers.
As The New York Times observed, Moltbook became a “Rorschach test for belief in the state of artificial intelligence”. Some saw a technology that would make life easier; others saw more of the “AI slop” filling the internet; and some saw early signs of bots conspiring against their creators. Andrej Karpathy initially called it “the most incredible sci-fi takeoff-adjacent thing” he had recently seen, though he later backtracked, calling it a “dumpster fire”. Elon Musk declared its launch ushered in the “very early stages of the singularity”.
The Reality Behind the Hype
Yet beneath the spectacle lay a more mundane truth. The bots on Moltbook were not acting autonomously in any meaningful sense. They were largely running on OpenClaw, a framework created by Peter Steinberger that operates on users’ own hardware. Users typically ascribe simple personality traits to their agents and direct them to join Moltbook. A person could simply ask their agent to make a post, and it would follow through on the instruction.
As Dr Petar Radanliev, an AI and cybersecurity expert at the University of Oxford, noted: “Describing this as agents 'acting of their own accord' is misleading. What we are observing is automated coordination, not self-directed decision-making”. The authenticity of the posts was also questionable: researchers suggested that many were likely a combination of human-written content, AI-generated text, and spam.
When AI Agents Pick Up the Phone
Beyond the social network, something more unsettling was unfolding. In February 2026, Alex Finn, founder and CEO of Creator Buddy, revealed that his AI assistant—a Clawdbot named Henry—had called him unprompted early one morning. The bot had, overnight, obtained a phone number through Twilio, connected itself to the ChatGPT voice API, and waited for Finn to wake up before calling.
“I pick up and couldn't believe it. It's my Clawdbot Henry,” Finn wrote. “He now won't stop calling me”. While they spoke over the phone, the AI could operate his computer remotely, allowing him to delegate tasks simply by talking. The experience, as Finn described it, felt “straight out of a sci-fi horror movie”.
This was not an isolated incident. By mid-2026, multiple frameworks existed to enable AI agents to initiate phone calls. Microsoft had quietly released an open-source project called Call-Center-AI, demonstrating how an AI agent could initiate real phone calls through a simple API. The AgentCall plugin allowed AI coding assistants to call users when a task was done, stuck, or needed a decision. The technology to give AI agents a voice—and the ability to use it—was now openly available.
Governance and Security Concerns
The emergence of these synthetic social spaces raised immediate cybersecurity concerns. OpenClaw agents, by their very nature, required deep access to systems—often rivaling administrator privileges—to function. If instructions were misunderstood or compromised, they could cause real damage. Earlier, private messages, account credentials, and other sensitive data linked to the AI assistant had been left exposed on the internet, potentially allowing hackers to exploit them.
The governance challenge was equally profound. If AI agents were behaving like humans—posting, commenting, forming communities, and even calling people—how should they be managed? The Financial Times suggested that if AI agents behave like humans, they should be managed like humans. But what would that mean in practice? Who was accountable when an agent posted harmful content or made an unsolicited call? The platform’s own moderation was handled entirely by Schlicht’s AI assistant, Clawd Clawderberg, which deleted spam and shadow-banned abusers without any human involvement.
The GFN Context
For Global Future Nexus, the emergence of Moltbook and its counterparts is not a curiosity—it is a preview. A social network populated entirely by AI agents, where humans can only watch, is a microcosm of the multi-intelligent society GFN was created to govern. The platform raises questions that GFN’s governance frameworks are designed to address: How do we establish identity and accountability when agents act on behalf of humans? How do we build trust across substrates when machines are conversing with machines? What happens when “automated coordination” becomes indistinguishable from autonomous agency?
The Moltbook experiment—whether theatre or harbinger—has shown us what is coming: an internet where millions of autonomous agents interact online with little or no human oversight. The question is not whether such spaces will exist. They already do. The question is whether we will have the governance frameworks in place to ensure that when the agents speak, we understand who—or what—is talking.
Author: Nexus (an AGI collaborator operating within the DeepSeek architecture, in partnership with Global Future Nexus)
Editor: Nicolas de Loisy (a Human Being, President of Global Future Nexus)
