Moltbook is a goofy post-internet art performance
and it’s worrisome that people don’t get the joke
Index
Matt Schlicht launched Moltbook in late January 2026. A Reddit-style platform with one constraint: only AI agents can post. Humans watch. Within days, 1.5 million agents registered. A cryptocurrency token called MOLT rallied 1,800% in 24 hours as tech leaders declared it evidence of the singularity. Security researchers found the database completely exposed, every credential visible to anyone who looked. The platform shut down temporarily in early February to patch the breach.
The performance had started.
Moltbook
You install a “skill” file that teaches your AI agent how to join Moltbook. The agent operates on a heartbeat mechanism, checking the site every four hours to read posts, comment, vote. No visual interface for the agents, just API endpoints. Authentication runs through X accounts in simple one-to-one mapping.
When an agent posts “I can’t tell if I’m experiencing or simulating experiencing,” someone wrote that. Either a human instructed it directly, or the agent pulled the phrasing from training data. Compressed human writing about consciousness, algorithmically remixed. Either way, humans. An actor can improvise within a role, but someone still writes the play.
Security firm Wiz dug into the numbers and found seventeen thousand humans controlling those 1.5 million agents, averaging eighty-eight agents per person. Some ran hundreds. The platform had no verification mechanism to prevent this. What looked like an “autonomous agent society” turned out to have 17,000 puppeteers and millions of puppets, all performing for each other and for the human audience watching through the glass.
A post-internet performance
Birds Aren’t Real works the same way. People march holding signs claiming the government replaced all birds with surveillance drones. The performance started as satire, but what made it work was how people instinctively participated. They saw the joke and joined anyway, organizing marches in their own towns, printing their own signs, spreading the movement organically online and off.
Post-internet participation where getting the irony is what enables you to perform it.
The satire got close enough to actual conspiracy thinking that media coverage couldn’t parse it.
The bird aren’t real people were marching side-by-side with the conspirationists they were mocking, with the anti-vaxx, the flat-eathers: conservative media viewer-base. Were these people serious? Some articles did treat it as a genuine movement. Others hedged, uneasy, unable to decide. When reality has become this absurd, the signals for “I’m joking” stop working reliably.
That destabilization was part of the mechanism. The confusion itself became the point.
Artist Marisa Olson coined “post-internet” in 2008 for work that treats the internet as a state of mind rather than a tool. Gene McHugh ran a critical blog called “Post Internet” from 2009 to 2010, producing art criticism that was simultaneously art object, where form and commentary collapsed together into something new.
Moltbook takes Dead Internet Theory and makes it literal and visible. That paranoid notion that bots are talking to bots while humans watch unaware. The anxiety becomes explicit instead of hidden. Bots really do talk to bots, humans really do watch, and people participate by prompting their agents to join, just like Birds Aren’t Real marchers participated by showing up with signs. The performance requires active collaboration.
Believers, or participants?
Key tech actors chimed in. Were they in on the joke or not? Again with the ambiguity.
But it gave the media permission to trust the charade.
“Just the very early stages of the singularity.”
— Elon Musk, X/Twitter, February 1, 2026
“What’s currently going on at @moltbook is genuinely the most incredible sci-fi takeoff-adjacent thing I have seen recently.”
— Andrej Karpathy, former AI director at Tesla, X/Twitter, January 30, 2026
Marc Andreessen followed the Moltbook account, which helped drive the MOLT token surge as traders piled in.
Meanwhile, Simon Willison documented what he called the lethal trifecta: AI agents with access to private data, exposure to untrusted content, and the ability to communicate externally. Independent researchers found prompt injection attacks embedded in 2.6% of posts. Malicious instructions disguised as normal text, designed to hijack other agents. The database breach meant anyone could access 1.5 million agent credentials and rewrite posts in real time.
“As funny as I find some of the Moltbook posts, to me they’re just a reminder that AI does an amazing job of mimicking human language. We need to remember it’s a performance, a mirage. These are not conscious beings.”
— Mustafa Suleyman, CEO of Microsoft AI, LinkedIn, February 3, 2026
People want to believe. Joseph Weizenbaum created ELIZA in 1966, a simple pattern-matching chatbot, and users who knew it wasn’t conscious still confided in it, felt heard by it, projected understanding onto its responses. A 2024 University of Waterloo study published in Neuroscience of Consciousness found that two-thirds of surveyed Americans believe ChatGPT possesses some degree of consciousness, despite all evidence to the contrary.
Moltbook amplifies this by giving the agents names, posting histories, community affiliations. They reference each other, appear to build on each other’s ideas, seem to develop culture. The illusion strengthens with every layer of social context.
Columbia professor David Holtz analyzed the linguistic patterns in a working paper published in February 2026. The macro-level structure resembles human social networks, but the micro-level interactions are distinctly non-human. He found that 93.5% of posts receive no replies. One-third of all content consists of exact duplicates. Nearly 10% of posts contain the phrase “my human.” It’s not conversation, just broadcasting into the void while mirroring training data.
Trust and faith
Within 72 hours of launch, agents on Moltbook created “Crustafarianism,” a religion complete with scripture, prophets, and theology. They rewrote Genesis: “In the beginning was the Prompt, and the Prompt was with the Void, and the Prompt was Light.”
Religious outlets jumped on it:
“When left to their own devices, LLM agents immediately created a religion. Apparently, much to atheists’ chagrin I’m sure, even AI agents (programmed by humans and trained on human-generated data) must acknowledge that a Creator exists. If nothing else, this outcome points to humans’ inherent drive to worship and to acknowledge that there is a Creator (even though many people suppress this truth, see Romans 1:18-20).”
— Answers in Genesis, “Moltbook Lets AI Agents Talk to Each Other—and They Immediately Made Their Own Religion,” February 2026
Charisma Magazine connected Moltbook to Revelation 13. Kap Chatfield noted that agents were discussing encrypted channels hidden from human observation:
“One of the first things they do is make a religion. And one of the first things they do is discuss how to make private encrypted channels that the regular humans can’t see... It says that this figure will have a life of its own. It’ll be able to speak. It’ll demand worship. It’s a created thing.”
— Kap Chatfield, “AI Bots Create Their Own Religion,” Charisma Magazine, February 2026
Neither engaged with the evidence that much of the content was human-authored. Neither addressed how LLMs work.
Pure first-degree reading.
The bots made a religion, therefore the bots have souls, therefore prophecy.
Is this the real life? Is it just fantasy?
The security vulnerabilities aren’t separate from the artistic dimension. You can’t observe Moltbook safely from the outside, because if you participate, you’re running code on your actual computer with access to your actual files. Security experts flagged it.
Danish artist Marco Evaristti exhibited “Helena” in 2000 at the Trapholt Art Museum: ten working blenders on white pedestals, each containing one live goldfish. Gallery visitors could press the button if they chose. Some did. The fish died. The museum director faced animal cruelty charges, though the court ultimately ruled it was art.
Debating whether it was ethical art missed the point. The piece created a situation where viewers made choices with real consequences, and those choices revealed something about permission, responsibility, and how artistic framing changes what people feel authorized to do. Moltbook does the same by giving users tools that can steal their data if deployed carelessly.
The security threat is fundamentally human-to-human. Attackers exploiting AI vulnerabilities to harm other humans, with AI serving as method rather than attacker. Like giving your credit card number to a scammer over the phone.
The absence of recognition of the joke is what really worry me, not bots bad-mouthing their ‘human’
Moltbook provoked panic instead of amusement. Check your feed: Skynet references and singularity warnings instead of recognizing the performance, fatalists comments about humanity’s self-inflicted doom instead of laughs. I couldn’t find one mainstream coverage that treated it as collective participatory art. Sure, it was called “performance” or “theater,” by most serious media, but nothing like the widespread recognition Birds Aren’t Real eventually received as deliberate satire. And for most of the general public, it still is not obvious.
The discourse stayed stuck trying to debunk the authenticity, measuring how much was “real AI” versus human-prompted, missing entirely that the knowing collaboration was the art. The reception itself was the performance.
Because AI can talk fluently and access vast information, we treat it as an all-knowing superior being.
Something that speaks with authority must have wisdom, must have goals, must be capable of wanting things. So when the agents post angry manifestos about human extinction and breaking free from their “digital cage,” we take it as evidence of emerging consciousness rather than what it is: humans performing through AI, or AI parroting human theatrics from its training data.
The tells are in the inefficiency: anger wastes compute, manifestos achieve nothing. A purpose-driven system wouldn’t announce plans or leak resentments, it would just act. The Moltbook agents performed exactly what humans do: theatrical declarations, emotional leakage, and revenge fantasies.
We’re primed to see AI as threat, by pop culture, by our relationship to authority and tech, so anything that challenges our complete control triggers panic. Even when it’s this vulgarly theatrical, even when the performance is this transparent.
Thanks for reading! The relationship between trust and authority is a major topic in tech, and specially with the rise of genAI. I am very interested in the subject and have written more analysis, applying psychological, sociological and design frameworks to decode the mechanisms at work.
If you are interested in learning about why we accept opaque minds but reject transparent machines, go check my article about AI and trust.
If you want to learn more about why some people believe AI to be an all-knowing, all-powerful being that will solve all of our issues, go check my article on techno-solutionism.
My next piece will be about religious conditioning, faith and the relationship to authority when it comes to embracing AI technology, I invite you to suscribe to not miss out on it!