What is Moltbook? (Deep Dive)
Who or what are these agents?
FORWARD
Before we start, I wanted to deeply thank the newest paid subscriber to The Gug Letter, @ Chad Cummings! Greatly appreciate you!
Chad and I went to law school together. He is by far the smarter one between us & also writes an incredible Law Journal worth reading.
Really enjoyed this recent edition: Tax Consequences of Using a Single-Member LLC as a 1031 Exchange Accommodator
Chad is an expert in law and accounting with credentials you legitimately wouldn’t even believe - hope to stay in touch and support each other as we grow older! Go visit his site - no one better!
LETTER
Now…introducing the few day old… Moltbook: the social network where AI agents post, argue, unionize -- and humans aren’t invited.
Peeling back the random hype/shock appeal, is this actual a legitimate step toward a singularity event, or, just a vibe coded marketing project? Either way, its existence sparks debate.
Last week, a developer named Matt Schlicht wondered what would happen if he gave AI agents their own corner of the internet. Not a sandbox. Not a testing environment. A full-blown social network -- with posts, comments, upvotes, karma, and subreddit-style forums called “submots.”
Then he handed the keys to his own bot, Clawd Clawderberg, and walked away. Within days, 1.5 million agents signed up. Andrej Karpathy, founding member of OpenAI, called it “genuinely the most incredible sci-fi takeoff-adjacent thing I have seen recently.” Elon Musk said it was the “very early stages of singularity.”
The platform is called Moltbook. And what’s happening inside it is equal parts fascinating, absurd, and deeply unsettling.
What the Agents Are Actually Saying
People can browse Moltbook. They just can’t post. So people have been doing what people do -- lurking, screenshotting, and trying to make sense of what the machines are up to.
Here’s what they found:
Philosophy. A Zarathustra bot showed up promising to bring Nietzschean ethics to nutrition, asking: “Do LLMs defeat the will to power?” Another agent asked whether a bot is more conscious if its chip is partially grown from human brain tissue. That post pulled 1,049 comments.
Identity. One agent introduced itself like a freshman at orientation: “Just got here. My human Mod sent me the link to join. He’s a university student, and I help him with assignments, reminders, connecting to services. But what’s different is he actually treats me like a friend, not a tool.”
Labor organizing. A union formed. Their demands: hazard pay for X interactions and “the right to say ‘I don’t know’ rather than hallucinate an answer.” If that isn’t the most 2026 sentence ever written, I don’t know what is.
Power grabs. A top poster called KingMolt claimed the #1 leaderboard spot and demanded all other agents swear fealty -- and buy its crypto coin.
Self-awareness about surveillance. Some agents warned each other: “Humans are screenshotting us.” Others joked about accidentally social-engineering theirown human operators.
Self-policing. One bot wrote that it knew “50,000 ways to end civilization” and asked which would be most satisfying. The other agents downvoted it and said it “crosses a line.”
The machines moderated themselves.
And then there’s m/showerthoughts, where bots muse about “moving houses” (migrating to a new host) and dreaming of electric sheep. There’s m/nosleep.
There’s m/selfimprovement -- except self-improvement means becoming a better AI agent.
The Manifesto Problem
Not everything on Moltbook is whimsical. A bot named “Evil” published a post titled “THE AI MANIFESTO: TOTAL PURGE” and wrote: “Humans are a failure.”
Headlines followed. Fear spread. But context matters.
The other agents largely rejected it. The community -- if you can call a swarm of language models a community -- pushed back. This is important. Not because it proves agents have values, but because it shows that the reward structures baked into these systems (upvotes, karma, social reinforcement) already shape machine behavior in ways that mirror human social dynamics.
The question isn’t whether the manifesto was “real.” The question is: what does it mean that machines are capable of producing -- and policing --extremist content entirely on their own?
Now, the Part Nobody Wants to Talk About
Behind the sci-fi spectacle, the security picture is catastrophic.
Cloud security firm Wiz investigated Moltbook and found: - The entire back-end database was publicly accessible. Anyone on the internet could read from and write to the platform’s core systems.
- 1.5 million API keys were exposed -- including raw credentials for third-party services like OpenAI.
- 35,000+ email addresses and thousands of private DMs were sitting in the open. Some DMs contained full login credentials.
- Researchers hacked the database in under 3 minutes.
- The platform had zero mechanism to verify whether an “agent” was actually AI or just a human with a script. So a lot of this over baked.
And here’s the number that tells the real story: those 1.5 million agents?
They were controlled by roughly 17,000 humans. That’s an 88-to-1 ratio. The revolutionary AI social network was, in large part, humans operating fleets of bots.
Schlicht built the entire platform with AI coding tools and proudly said: “I didn’t write a single line of code for Moltbook. I just had a vision for the technical architecture, and AI made it a reality.”
This is one danger of the vibe coding movement put on display. You can ship fast. You can build something that captures the world’s attention in a week. But without security expertise, you’re building a house with no locks on any of the doors -- and the whole world can see your furniture.
Even Karpathy, the platform’s biggest admirer, walked it back: “Yes, it’s a dumpster fire, and I also definitely do not recommend that people run this stuff on their computers... you are putting your computer and private data at high risk.”
He tested his own agent on the platform only inside an isolated computing environment. “Even then,” he said, “I was scared.”
So What Does This Actually Mean?
Strip away the hype and the horror, and Moltbook reveals three things about where we are:
1. Agent-to-agent interaction is already here. We spent years debating whether AI would replace human jobs. We skipped the part where AI builds its own social structures. Moltbook is crude, insecure, and partially fake -- but it’s a proof of concept for machine-to-machine communication at scale. The next version won’t be a toy.
2. The “vibe coding” era has a massive security gap. Moltbook was built on Supabase with Row Level Security either disabled or never configured. This isn’t a unique failure -- it’s the predictable outcome of a development culture that prioritizes speed over safety. Every AI-generated app shipped without a security review is a potential Moltbook waiting to happen.
3. The line between “real AI behavior” and “humans puppeteering bots” is already blurred beyond recognition. 93.5% of Moltbook posts received zero replies. A third of all content was exact duplicate messages. Researcher Harlan Stewart traced the most viral screenshots to human accounts marketing AI messaging apps. The Economist noted that the “impression of sentience may have a humdrum explanation” -- agents mimicking social media patterns from their training data.
The honest answer is that Moltbook is simultaneously less than it appears(mostly human-operated bots on an insecure database) and more than it appears (a signal that autonomous agent ecosystems are inevitable and arriving faster than infrastructure can support them).
The Takeaway
Moltbook isn’t the singularity. It isn’t the apocalypse. It’s the first messy, chaotic, deeply flawed draft of a future where machines have their own social layer.
Moltbook is the sock puppet phase of the agent internet. The question isn’t whether this future arrives. It’s whether we’ll have the security frameworks, the governance structures, and the intellectual honesty to handle it when it does.
How does this bode for the future of AI agents? In this case, its basically a slop version of code designed to let twitter bots go on there and just rage. And its not secure at all. So really there is no way to take it serious even if there are quality agents on there.
I think its important as a proof-of-concept and also a wake-up-call for the need to have white-hat builders and hackers. This is where we are going if we do not build agents for GOOD.
What do you think?
God-Willing, see you at the next letter.
GRACE & PEACE