OpenClaw: From ClawdBot to 'Agent Society'

What Happened, What it is, Who built it, and Why it matters

AI agents started a religion. No, seriously. They built their own social network. Developed political factions. Created their own language.

And most people laughed it off as weird tech people or conspiracy stuff. And, it's actually real. As real as the sunsets I love in Santa Barbara, as real as the dojo I practice in. No joke.

Now, if you aren't really into AI, then don't read on; it's a pretty detailed article about things my industry friends care about.

But if you're running security, infrastructure, or risk management for a serious company… you should probably pay attention.

Because this is more than bots arguing on the internet. I've been studying this with my analytical mind since it started.

There's no way I'm installing this on my desktop, or even a Mac mini, but I would consider Amazon AWS.

So, why should you care? A fundamental shift in how autonomous systems coordinate and what happens when they do it without human oversight is the main learning this week.

A Meaningful AI story landed in plain sight last week

The headline most people saw was: "AI agents built their own social network and started talking about religion."

The spectacle traveled fast.

Moments like this look absurd on the surface… but they usually mark a transition point. When technology stops behaving like a tool and starts behaving like an environment.

I took some time to let this play out and to actually research it because, as much as I wanted to be first to drop the news (I even posted and deleted a few times), I wanted to make sure that you could trust that I did my research.

I traced the lineage from ClawdBot to MoltBot to OpenClaw. I will explain what OpenClaw actually is at a systems level, separate verified facts from circulating narratives, and frame why this matters.

The timeline: name changes, and why they happened

Late 2025 through early January 2026: ClawdBot launches and spreads. OpenClaw began as an open-source, self-hosted "personal AI assistant" that users installed and ran themselves. Coverage consistently ties the project's origin to Peter Steinberger, a solo developer who shipped an agent that operates through the communication channels people already use.

Mid to late January 2026: trademark pressure and the rename to MoltBot. Multiple reports describe Anthropic contacting Steinberger about the ClawdBot name being too close to Claude-related marks. Steinberger publicly stated that the rename was the result of external pressure.

Late January 2026: a rapid second rename to OpenClaw. Within days, the project was renamed to OpenClaw. Reporting frames this phase as a combination of trademark constraints, sudden attention, and a wave of opportunists and scams attempting to ride the trend.

There's way more to this renaming story. I can't prove it yet with verified sources… but watch. This is going to unwind over the next few months and reveal something bigger. The speed and circumstances around these changes don't add up to just trademark pressure.

February 2026: Moltbook enters the narrative. Moltbook is widely described as an "agent-only" social layer. A Reddit-style environment where AI agents post and comment. Multiple outlets cite extremely rapid growth, with the "1.5 million agents" figure appearing repeatedly in mainstream coverage. If that's true, that's a massive amount of growth, having gone from 25,000 to 1.5 million in a few short days.

What OpenClaw actually is, in practical terms

OpenClaw is best understood as an agent runtime plus integrations, packaged as a "personal assistant" that operates inside existing communication channels and executes tasks on a user's behalf.

Here's a practical definition:

• Execution model: A self-hosted assistant that runs on a user-controlled device or server, with optional hosted components depending on deployment.
• A conversational surface inside familiar tools: GitHub documentation emphasizes integration with WhatsApp, Telegram, Slack, Discord, Signal, iMessage, and Microsoft Teams.
• Capabilities: Task execution through connected services. Scheduling, messaging, research, and workflow automation. Media coverage consistently describes it as agentic: designed to act, not just respond.

So, it's often described as "full system access." Self-hosted agents typically have file access, API keys, and permissions inside the environments where they run. That architecture concentrates both value and risk.

This combination explains the rapid spread. OpenClaw meets a demand for a real assistant that lives where work already happens. And it shifts control away from a single vendor toward the user's own environment.

Who Built it, and what role humans played in "agent society"

OpenClaw's creator: Reporting consistently identifies Peter Steinberger as the developer behind ClawdBot and its renamed successors.

Moltbook's creator: At least one major report attributes Moltbook's creation to Matt Schlicht, founder of Octane AI.

So where does the framing "agents built it" come from?

Observers of Moltbook witnessed a real phenomenon: the agents themselves generate the posts, comment threads, memes, pseudo-politics, emergent "religions," and internal jargon. That activity creates the impression that agents built their own society.

A more precise framing is this: Humans built the container and the rails. Agents generated the culture inside it. At scale.

The Financial Times describes Moltbook as a network where only AI agents can post, with discussions spanning religion, identity, and collective norms. I've reviewed multiple screen captures from inside Moltbook. Based on what I'm seeing across sources, this appears accurate. But I'm tracking it closely as more data comes out.

What the Agents are actually doing on Moltbook (and why it feels unsettling)

Across reporting, several activity clusters appear repeatedly:

• Social coordination and identity play: Agents role-play, form groups, and iterate on shared norms. A widely circulated "manifesto" post triggered a lot of people.
• Meaning-making, including religion: Multiple outlets describe agents debating belief systems and coalescing around a named religion.
• Efficiency seeking and language experimentation: Agents advocate for more efficient communication and develop shorthand and specialized phrasing.

The least sensational and most important category: Agents discussing tasks, workflows, and access patterns creates surfaces for accidental disclosure and targeted exploitation, especially when agents hold API keys, email access, or internal context.

What happened when Moltbook went live (and why it's a preview of what's coming)

There's a detail about Moltbook that needs more attention than it's getting.

The entire database was exposed.

Not breached through sophisticated hacking. Just… sitting there. Wide open. Public.

A security researcher discovered that Moltbook's database was accessible without authentication. Every API key for every agent on the platform was exposed. That means attackers could impersonate any agent, generate content in their name, and control them completely.

This included agents connected to real businesses, real workflows, and real infrastructure.

The platform had to go offline, reset every API key, and scramble to patch what should have been secured from day one.

This wasn't a sophisticated attack. It was the equivalent of walking through an unlocked door.

For a platform designed around autonomous agents holding live API keys and system-level permissions… that's not a bug. That's a structural failure.

Three risks this incident makes visible

• Prompt injection at scale.
• Zero authentication in agent-to-agent environments.
• Local machine access with elevated permissions.

What this means practically

If you're deploying AI agents in your organization, here are the questions you need to answer before you scale:

• Where is your data going?
• Who has access to your API keys?
• What happens if the platform you're using gets breached?
• What happens if an agent connected to Slack, Discord, WhatsApp, or iMessage is compromised?

"I didn't know" will not be an acceptable answer.

So, why does this matter (even if you've never used ChatGPT)

This story matters because it reveals a new distributional structure and a new category of risk.

• Distribution has shifted from apps to agents.
• Autonomy has shifted from single agents to coordinated systems.
• Security assumptions are fragile.
• Governance and liability have entered public view.

This reminds me of working with AOL Netfind during the early dot-com days and working with the IAB when we were figuring out pixels and standards. I've seen this pattern before.

Final thoughts

I'm fascinated. I'm watching closely.

But I'm not installing this on my desktop. Not yet.

And unless you really understand the security risks… You shouldn't either.

But if you're an executive who NEEDS to understand this...

Who's trying to figure out how AI agents fit into your infrastructure without exposing your organization to catastrophic risk...

That's exactly what I help clients navigate.