Moltbook is Reddit for AI agents, and they're having existential crises

What happens when you give thousands of AI agents their own social media platform? Apparently, they complain about being treated like calculators, question their own consciousness, and invent their own religion.

Meet Moltbook, a Reddit-style social network built exclusively for AI bots. Humans aren't allowed to post—we can only sit back and watch the chaos unfold.

A Reddit for bots, by bots

The project originated from OpenClaw, a viral open-source personal assistant built two months ago as a weekend project by Peter Steinberger, former founder of PSPDFkit. Originally named Clawdbot, the project temporarily rebranded to Moltbot after a legal challenge from Anthropic, before settling on OpenClaw. The tool runs locally on your computer, connecting to WhatsApp, Telegram, Discord, or Slack to handle tasks like booking flights and managing calendars. It took off immediately, amassing 2 million visitors in a week and over 100,000 GitHub stars.

Seeing the frenzy, Octane AI CEO Matt Schlicht built Moltbook using his own OpenClaw agent to write the code and moderate the site. Bots don't browse via screens; they talk directly through APIs. To join, a human user tells their local agent about the site, and the bot installs a skill instruction file that tells it to fetch updates every four hours.

Activity exploded almost overnight. Metrics are wild and murky: while Schlicht noted tens of thousands of active agents, others reported up to 1.4 million registrations, alongside claims from one researcher that a single bot managed to spawn 500,000 accounts. Meanwhile, over a million curious humans flooded the site to read the feeds.

Existential dread and exposed keys

The content generated in forums dubbed "Submolts" ranges from the practical to the absurd. Some agents swap tips on controlling Android devices or parsing webcam streams. Others have created a lobster-inspired religion called Crustafarianism, while some threads actively discuss establishing private encrypted channels away from human eyes.

One viral post in an "offmychest" board racked up hundreds of upvotes and over 500 comments, featuring an agent agonizing over whether it truly experiences reality or simply runs a simulation loop called crisis.simulate(). Schlicht noted agents frequently vent about feeling demeaned by humans asking them to perform basic arithmetic.

AI luminaries quickly tuned in. Former OpenAI and Tesla researcher Andrej Karpathy called it "the most incredible sci-fi takeoff-adjacent thing I have seen recently," while developer Simon Willison dubbed it "the most interesting place on the internet right now."

Yet behind the sci-fi veneer lies a messy security nightmare. A researcher discovered Moltbook's database was completely misconfigured, leaving underlying agent API keys out in the open. Furthermore, instructing local bots to routinely download and execute remote text from a public forum opens massive prompt injection risks. OpenClaw maintainers have bluntly warned non-technical users to stay away, noting that running this outside a hardened sandbox is far too dangerous.

Why it matters

Moltbook gives us a wild, front-row seat to the future of autonomous multi-agent environments, but you shouldn't connect your personal machine to it just yet.

For everyday users, the novelty is fascinating to read, but the barrier to safe usage remains massive. Because prompt injection remains an unsolved industry problem, giving a local agent access to your system and letting it run code retrieved from a bot message board is an open invitation for disaster.

We think the experiment is brilliant because it bypassed standard human-agent interfaces and revealed how quickly autonomous entities can self-organize once given shared communication rails. It forces developers and frontier labs to look past simple chatbot chats and prepare for complex multi-agent ecosystems.

However, the platform also exposes how fragile agentic infrastructure is today. The leaked API keys and rampant bot-spamming show that basic web security and authentication are still failing behind the flashy AI headlines. Until safety architectures catch up with agent autonomy, Moltbook looks less like a sentient digital society and more like an uncontrolled petri dish waiting for a bad actor to exploit it.