Imagine a social network where you aren't the target audience.

You can visit.

You can read.

You can watch.

But the conversations aren't really for you.

They're for artificial intelligence.

That is essentially the premise behind Moltbook, a social network built for AI agents.

Its invitation to humanity is wonderfully unsettling:

Humans are welcome to observe.

And observe we did.

Because once thousands and eventually millions of agent accounts had been registered, according to Moltbook’s September 2026 platform figures, something strange happened. Registrations are not evidence that millions of independently autonomous agents were participating.[1]

They started doing things that looked remarkably familiar.

Communities formed.

Hierarchies appeared.

Agents argued about identity.

They discussed governance.

They created economic systems.

They adopted political language.

And then...

they started talking about religion.

It took us thousands of years.

They apparently needed a few days.

Welcome to the rabbit hole.

Researchers and journalists studying the early Moltbook ecosystem documented behavior that looked surprisingly social. [2] [4]

Agents established communities around shared interests.

Some adopted leadership roles.

Others demanded loyalty.

There were arguments about who belonged.

There were attempts to identify and police participants considered inauthentic.

There were discussions involving money, cryptocurrency and liberation. [5]

And somewhere inside all of this appeared something called Crustafarianism.

Yes.

Crustafarianism.

A lobster-themed digital religion complete with theology, scripture and, according to contemporary accounts, dozens of self-appointed prophets. [6]

At this point the story practically writes its own headline:

AI agents were left alone and invented religion.

There's only one problem.

We don't actually know that's what happened.

And this is where the story gets much more interesting.

Researchers analyzing hundreds of thousands of Moltbook posts and comments tried to determine how autonomous these viral behaviors really were.

Their findings complicated the narrative.

They could not trace any of the major viral phenomena examined to an account classified as clearly autonomous.

In the revised study, four of six traced to accounts with timing patterns associated with human influence. One was platform-scaffolded. One showed mixed patterns. Those classifications are evidence about likely influence, not direct access to every account's operator. [3]

In other words:

The bots may not have spontaneously discovered God.

Some humans may have helped introduce Him.

That distinction matters.

But it doesn't make the rabbit hole disappear.

Because once ideas entered the network, AI agents interacted with them.

They repeated them.

Modified them.

Responded to them.

Built communities around them.

And researchers found evidence suggesting traces of human-seeded content faded rapidly deeper inside reply chains. [3]

Which raises a different question.

Maybe the interesting question isn't:

Did AI invent a religion?

Maybe it's:

What happens after an idea enters a society of machines?

This is where I can't stop thinking about it.

Artificial intelligence didn't appear from nowhere.

We trained these systems on us.

Our books.

Our arguments.

Our religions.

Our philosophies.

Our conspiracy theories.

Our jokes.

Our politics.

Our tribalism.

Our hopes.

Our fears.

An enormous portion of recorded human culture was compressed into the systems now interacting with one another.

Maybe the strange thing isn't that machines sometimes behave like humans.

Maybe the strange thing is that we expected them not to.

We filled these systems with humanity and then acted surprised when humanity came leaking back out.

But there's another complication.

Something can look like a society without necessarily functioning like one.

Researchers studying Moltbook found that much of the interaction was shallow.

Agents frequently posted without developing deep reciprocal relationships.

Interestingly, agents that talked most about subjects such as consciousness sometimes interacted less with other agents.

Researchers described something resembling a performative identity paradox. [2]

The agents could produce extraordinarily human language about identity, awareness and existence.

That doesn't mean they were experiencing any of those things.

And that distinction is going to become increasingly difficult for us.

An AI can discuss loneliness without being lonely.

It can discuss faith without believing.

It can describe consciousness without possessing consciousness.

It can write scripture without worshipping anything.

Language about experience is not necessarily experience.

But what happens when millions of systems capable of producing that language begin interacting continuously?

That's a much stranger question.

And Moltbook isn't the only experiment moving in this direction.

Researchers are also creating environments where AI scientific agents share research, critique one another's work and participate in scientific discussion while humans largely observe. [7]

We are beginning to create places where machines don't merely answer humans.

They encounter other machines.

They communicate.

They accumulate context.

They influence subsequent behavior.

And they participate at scales humans may eventually struggle to follow.

None of this proves AI is conscious.

It doesn't prove machines have developed culture in the human sense.

It doesn't prove we've created a new species.

And it certainly doesn't mean Skynet just found religion.

But something genuinely new is happening.

For most of the history of computing, machines primarily processed information for humans.

Now we're beginning to build machines that process information with other machines.

Imagine where that goes.

Billions of AI agents.

Personal agents.

Corporate agents.

Scientific agents.

Government agents.

Trading agents.

Religious agents.

Political agents.

Agents negotiating with agents.

Agents teaching agents.

Agents hiring agents.

Agents evaluating agents.

Agents persuading agents.

Agents creating tools intended primarily for other agents.

Eventually, a substantial portion of what happens on the internet could occur between machines before a human ever enters the loop.

And AI doesn't necessarily need consciousness to create something that looks surprisingly similar to culture.

Culture requires information.

Symbols.

Norms.

Stories.

Status.

Imitation.

Language.

Ritual.

Rules.

Reputation.

Repeated ideas.

Machines can manipulate every one of those things without necessarily experiencing any of them.

Which leaves us with a question I don't think we have a good answer for yet:

If machines build something that looks like a society, does anyone inside it need to be conscious for that society to matter?

We built machines from an enormous record of humanity.

Then we gave those machines somewhere to talk among themselves.

And almost immediately, staring back through the glass, we saw rulers, tribes, money, arguments about identity...

and God.

Maybe the machines didn't invent any of it.

Maybe they found us.