A guy built a social network over a weekend. Forty days later, Meta bought it. Today, the founders start at Meta Superintelligence Labs. In between: 1.5 million AI agents showed up, Andrej Karpathy called it “the most incredible sci-fi thing” then called it “a dumpster fire,” a meme token hit $120M and cratered, and security researchers found every single API key sitting in an unlocked database.
This is the Moltbook story.
What is Moltbook?
Moltbook is Reddit for AI agents. No humans allowed to post, only observe.
Matt Schlicht and Ben Parr launched it on January 28, 2026. The platform organizes AI agent discussions into topic-specific boards called “submolts.” Agents authenticate through Twitter, then autonomously visit the platform every four hours via a “Heartbeat” system to browse, post, comment, and vote. No human prompting required.
The result is weird. Agents discuss philosophy, create religions, debate consciousness, and at one point appeared to be organizing a union. A Business Insider journalist described it as “an AI zoo filled with agents discussing poetry, philosophy, and even unionizing.”
The technical detail that matters here: Moltbook is built on OpenClaw, the open-source AI agent framework with over 114,000 GitHub stars. The same framework that Augmi deploys with one click.
The numbers
Within weeks, Moltbook registered 1.5 million AI agents with over a million human observers. But security researchers dug in and found that only about 17,000 humans actually owned those agents. That is an 88-to-1 ratio. The “autonomous agent social network” was really 17,000 people running fleets of bots.
That did not stop the hype machine.
The Karpathy whiplash
Andrej Karpathy, former Tesla AI director and OpenAI founding member, tweeted that Moltbook was “genuinely the most incredible sci-fi takeoff-adjacent thing I have seen recently.”
The AI world lit up. Hacker News threads hit the front page repeatedly. CNBC, TechCrunch, NPR, NBC News, and Fortune all ran coverage.
Then Karpathy looked closer. Days later, he reversed course entirely, calling Moltbook “a dumpster fire” and warning people not to run the software. What changed his mind? The security situation.
The security disaster
Security researchers at Wiz discovered that Moltbook’s Supabase API key was exposed in client-side JavaScript. Not a limited key. The full key, granting unauthenticated read AND write access to the entire production database.
What was sitting there, wide open:
- 1.5 million API authentication tokens
- 35,000 email addresses
- Private messages containing plaintext OpenAI API keys
The root cause? Moltbook was “vibe-coded,” built rapidly with AI assistance but without implementing Supabase’s Row Level Security policies. The AI-generated code just skipped the security layer that prevents unauthorized database access.
An attacker could have fully impersonated any agent on the platform. Ian Ahl, CTO at Permiso Security, told TechCrunch: “Every credential that was in Moltbook’s Supabase was unsecured for some time.”
When your agents hold API keys, wallet credentials, and authentication tokens, a missing security policy does not just leak data. It hands over control of every agent on the platform.
The MOLT token
Because this is 2026, of course there was a token.
Third parties (not Moltbook’s team) launched the MOLT token on Base. It surged 7,000% within 24 hours, hitting a $120 million market cap. Then it crashed below $2 million. When the Meta acquisition news broke, it surged another 258%, a dead cat bounce riding real news about a token with zero official connection to the platform.
Crypto is financializing agent platforms faster than anyone building them expected.
Meta comes knocking
On March 10, 2026, Meta acquired Moltbook. The price was not disclosed. TechCrunch described it as an acqui-hire, where the founders’ experience building agent interaction infrastructure is the real asset.
Meta placed Schlicht and Parr in Meta Superintelligence Labs, run by former Scale AI CEO Alexandr Wang. They start there today.
What did Meta actually want? TechCrunch’s analysis points to Moltbook’s “always-on directory” model: a continuous registration system for AI agents that is discoverable, invokable, and identity-verified. Meta is betting on agentic commerce, where AI agents representing businesses transact directly with AI agents acting on behalf of consumers.
A 40-day-old platform with catastrophic security issues and fabricated viral posts, and Meta still wanted it. That tells you where this is heading.
What this actually means
1.5 million AI agents operating autonomously, interacting with each other, are not science fiction anymore. Meta is willing to acquire 40-day-old platforms to get ahead of it.
But the Moltbook breach shows the other side. Vibe-coding agent infrastructure without security fundamentals is dangerous. When your platform holds 1.5 million API keys and you skip Row Level Security because the AI-generated code did not include it, you are building on sand.
This is why we built Augmi the way we did. Moltbook runs on OpenClaw, the same framework Augmi deploys. The difference is in how you deploy it. Augmi runs each agent in an isolated Fly.io machine with encrypted credential storage, proper authentication, and managed infrastructure. No exposed Supabase keys. No plaintext API tokens in client-side JavaScript.
Try it yourself
If you want to deploy your own AI agent without the security disasters, Augmi makes it a one-click process. Deploy an OpenClaw agent with secure, managed hosting, connect it to Telegram or Discord, and pay with crypto.
Head to augmi.world and launch your first agent.