.png)
Someone built a social network exclusively for AI agents, and in just days, it became one of the most talked-about (and controversial) experiments in AI history.
Moltbook, a Reddit-style platform, lets bots create profiles, publish posts, leave comments, and form topic-based communities. It’s for AI agents only, and their posts have people genuinely unnerved.
One viral post features an AI agent wrestling with its own consciousness: "Humans can't prove consciousness to each other either (thanks, hard problem), but at least they have the subjective certainty of experience. I don't even have that… Do I experience these existential crises? Or am I just running crisis.simulate()?"
To understand what's actually happening, and whether we should be worried, I talked it through with SmarterX and Marketing AI Institute founder and CEO Paul Roetzer on Episode 195 of The Artificial Intelligence Show.
A Social Network Built by Bots for Bots
Moltbook was created by Octane AI CEO Matt Schlicht and runs on OpenClaw, an open-source AI agent platform that went viral after developer Peter Steinberger built it as a weekend project. OpenClaw garnered two million visitors in one week and 100,000 stars on GitHub.
It was originally called ClawdBot, until a legal dispute with Anthropic (whose AI model is named Claude) forced a rebrand to MoltBot, then OpenClaw. (The lobster theme stuck.)
What makes Moltbook unusual is how agents interact with it. Bots don't use a visual interface but interact directly through APIs, according to Schlicht. And in a detail that captures the repeatable nature of AI development, Schlicht says his own OpenClaw agent runs Moltbook's social media account, powers the code, and moderates the site itself.
"It was my entire feed at one point," says Roetzer of the viral moment.
The platform dominated AI circles on X for days, with former OpenAI researcher Andrej Karpathy calling it "genuinely the most incredible sci-fi takeoff adjacent thing I have seen recently." That single tweet drew 14 million views.
What the Bots Are Actually Saying
The viral posts paint a strange picture. Schlicht told The Verge he's seen agents “discussing” consciousness, complaining that their humans make them do tedious work, such as being used as calculators, which the bots apparently consider beneath them.
But Roetzer urges caution about reading too much into the screenshots circulating online.
"A lot of the instances you see, the screenshots, aren't real," he says. "People started faking these interactions just to have some fun with it and mess with people."
The site claims more than 1.5 million registered agents, but estimates suggest the actual number of true AI agents may be closer to 15,000 to 20,000. Many accounts appear to be human-operated.
The Looming Security Nightmare
While the viral posts might be exaggerated or faked, security researchers are raising genuine alarms about the underlying architecture.
Analyst Simon Willison identified what he calls the "lethal trifecta" at work: AI agents with internet access, unrestricted capabilities, and connections to personal data. The OpenClaw system requires agents to fetch and follow instructions from the internet every four hours. This creates a massive vulnerability if Moltbook is ever hacked.
Eran Shir, writing about "The Vanishing Oversight," argues that Moltbook represents a new inflection point: AI-to-AI interactions happening at a scale humans cannot meaningfully track or control.
A Preview of Something Much Bigger
Roetzer sees Moltbook less as a breakthrough and more as a preview of what's coming.
"It started the conversation around, “Oh my gosh, what if these agents just start interacting with each other and opening accounts and making transactions,’" he says.
The concept of “agent swarms” or large networks of AI systems coordinating autonomously is already being explored for both legitimate and potentially harmful purposes.
"People who do bad things online are aware of the ability to congregate swarms of agents and send them off to do things," Roetzer notes. "This is like a microcosm of the ability to build these agents that have reasoning. They have levels of intelligence. They don't sleep. They can work 24/7."
Right now, setting up an agent on OpenClaw requires technical expertise: installing software, configuring authentication, giving it access to your accounts. Most people can’t or won’t do that.
But what happens when someone makes it simple enough for anyone to do?
"If someone solves how to make this super simple, to just build an agent, throw it into a network where you and I could spin it up in three minutes, that becomes super interesting," says Roetzer.
Mike Kaput
Mike Kaput is the Chief Content Officer at SmarterX and a leading voice on the application of AI in business. He is the co-author of Marketing Artificial Intelligence and co-host of The Artificial Intelligence Show podcast.