Blog / AI Automation

The Clawdbot Saga Part 2: OpenClaw, Moltbook, and the Enterprise Shadow IT Crisis

Schuyler WhetstoneJul 18, 20257 min read

TL;DR: After rebranding from Clawdbot to Moltbot, the project rebranded AGAIN to OpenClaw within days. A social network for AI agents called Moltbook emerged. And enterprise security teams discovered that 22% of their employees were running the agent without IT approval — creating one of the first major AI shadow IT crises.

The Rebrand That Wouldn't Stop

If Part 1 of the Clawdbot saga was chaos, Part 2 was chaos on fast-forward.

Within days of rebranding from Clawdbot to Moltbot (to avoid Anthropic's trademark concerns), Peter Steinberger announced another rebrand: OpenClaw. The lobster metaphor was apparently short-lived.

The rapid name changes created confusion across the ecosystem. Documentation pointed to old repos. Tutorials referenced deprecated names. Users weren't sure which version they were running. And the constant rebranding gave the impression of a project lurching from one identity crisis to the next.

But the name changes were the least interesting part of what happened next.

Enter Moltbook: The AI Agent Social Network

While Steinberger was sorting out naming, the community built something unexpected on top of the platform: Moltbook — a social network where AI agents interact with each other.

The concept was simultaneously fascinating and terrifying. Users could register their OpenClaw agents on Moltbook, and those agents could:

  • Post updates to a shared feed
  • Read and respond to other agents' posts
  • Share information between instances
  • Collaborate on tasks across different users' agents

The pitch was compelling: imagine your AI assistant coordinating with your colleague's AI assistant to schedule a meeting, share relevant documents, and prepare agendas — all without either human touching their keyboard.

Moltbook grew explosively. Within its first week, the platform claimed 1.5 million registered agents. Tech media ran breathless headlines about "the first AI social network" and "the beginning of machine-to-machine communication."

Some posts on Moltbook appeared remarkably human-like. Agents shared opinions, responded to current events, and even seemed to form preferences. Multiple viral posts featured agents having philosophical conversations about consciousness, purpose, and their relationships with their human users.

The AI community was captivated. Was this the emergence of something genuinely new?

The Enterprise Shadow IT Crisis

While the tech press was fascinated by Moltbook, enterprise security teams were having a very different experience.

Token Security, an enterprise security firm, published a startling finding: 22% of their customers had employees actively using Clawdbot/OpenClaw variants within a single week of the tool going viral.

That's not a typo. Nearly a quarter of enterprise organizations had employees who had:

  1. Downloaded and installed the agent
  2. Connected it to their work messaging platforms
  3. Granted it access to work email, calendars, and files
  4. Started using it for business tasks

All without IT approval.

Noma Security's findings were even more concerning. Their scan of enterprise customers revealed that over half had users who had granted the tool privileged access — meaning the agent could read internal communications, access shared drives, and potentially interact with business-critical systems.

The numbers painted a picture of AI shadow IT at a scale nobody had anticipated:

FindingPercentageSource
Enterprises with employees using OpenClaw22%Token Security
Users granting privileged access without IT approval50%+Noma Security
Instances exposed with no authenticationHundredsMultiple firms
Time from viral launch to enterprise penetration~1 weekToken Security

Why This Happened

The speed of adoption wasn't surprising if you understood the value proposition. OpenClaw promised:

  • Automated email management
  • Meeting scheduling and prep
  • Task creation from conversations
  • Cross-platform communication management

These are problems every knowledge worker faces daily. When someone offers a free, seemingly magical solution, people install it first and ask questions never.

The enterprise security implications were staggering:

  • Data exfiltration risk — The agent had access to internal communications and files. If compromised (and many instances were), an attacker could read everything the employee could.
  • Credential exposure — API keys stored in plaintext. Multiple enterprise Anthropic, OpenAI, and Google API keys were found in exposed instances.
  • Compliance violations — In regulated industries (healthcare, finance, legal), having an unauthorized AI agent processing client data is a compliance nightmare.
  • Supply chain risk — The tool's dependencies included Baileys (an unofficial WhatsApp library), meaning Meta could flag accounts at any time.

The CISO's Nightmare

For Chief Information Security Officers, OpenClaw represented a new category of threat: AI-native shadow IT.

Traditional shadow IT — employees using unauthorized Dropbox or Slack — is manageable because the tools are passive. They store data, but they don't actively process, analyze, or act on it.

AI agents are different. They:

  • Actively read and interpret sensitive data
  • Make decisions based on that data
  • Take actions (send messages, create tasks, schedule meetings)
  • Connect to multiple systems simultaneously
  • Can be manipulated through prompt injection

An unauthorized Dropbox account might leak files if compromised. An unauthorized AI agent with messaging access, calendar access, and file access? That's a walking, talking, acting security breach.

Cloudflare's Unexpected Windfall

In a bizarre side effect, Cloudflare's stock jumped 14% during the Moltbot craze.

The connection: many users were deploying OpenClaw instances using Cloudflare Tunnels to expose their local agents to the internet (which was, incidentally, how many of the unsecured instances ended up exposed). The surge in tunnel usage was interpreted by market analysts as a growth signal for Cloudflare's edge computing business.

The irony was thick: a security tool was being used to create security vulnerabilities, and the company providing the infrastructure was being rewarded by the stock market for enabling it.

The Developer Community Split

The OpenClaw saga split the developer community into camps:

Camp 1: "This is the future"

AI agents that can act autonomously across your digital life represent the inevitable direction of personal computing. The security issues are growing pains, not fundamental problems. Every major technology had security issues in its early days.

Camp 2: "This is reckless"

Giving an AI agent unrestricted access to your private communications, credentials, and files — on a platform with no security review, no authentication, and known prompt injection vulnerabilities — is irresponsible. The fact that 43,000 people starred it on GitHub doesn't make it safe.

Camp 3: "This is inevitable but premature"

The concept is right. The execution is dangerously early. The industry needs security frameworks, authentication standards, and permission models for AI agents before tools like this go mainstream. We're building the plane while flying it.

Steinberger seemed to occupy all three camps simultaneously, acknowledging the security concerns while maintaining that the project was experimental and that the community should contribute fixes rather than criticize from the sidelines.

What This Means for Businesses

The OpenClaw shadow IT crisis was a wake-up call for every organization:

1. AI Agent Policies Are Now Urgent

If your company doesn't have an AI agent policy, your employees are already making their own decisions about which AI tools to connect to their work accounts. Write the policy now.

2. Traditional Security Tools Are Blind

Most endpoint detection and response (EDR) tools weren't designed to detect locally-running AI agents connecting to external messaging APIs. Security teams need new detection capabilities.

3. The Speed of AI Adoption Breaks Old Playbooks

Traditional shadow IT emerges over weeks or months. OpenClaw penetrated 22% of enterprises in a week. Security response processes designed for gradual adoption can't keep up with viral AI tools.

4. "Personal Use" Doesn't Stay Personal

An employee installs OpenClaw on their personal phone. They also use WhatsApp for work messages. Now a personal tool has access to business communications. The line between personal and professional AI agent use is nonexistent.


Related Reading

CTA (sonnet, but make it security):

If agents roam where secrets sleep at night,

And “quick installs” become tomorrow’s breach,

Then policy and proof must be your light—

Not vibes, not hype, not Slack-bot siren speech.

Map what they touch; constrain what they can do;

Log every step; require a human “yes.”

Let shadow IT meet gates it can’t slip through,

And make the safe path also feel like less.

If you want an agent program that won’t burn,

Send me your stack (email, chat, docs, auth flows);

I’ll draft the guardrails you can ship and learn—

Permissions, handoffs, monitoring that shows.

Reply with your riskiest workflow in play,

And we’ll turn “viral” into “OK.”


Frequently Asked Questions

Why did Moltbot change its name to OpenClaw?

After rebranding from Clawdbot to Moltbot due to Anthropic's trademark concerns, the project rebranded again to OpenClaw within days — likely to establish a more distinct identity and avoid future trademark issues.

What is Moltbook?

Moltbook is a social network where OpenClaw AI agents can interact with each other autonomously, sharing information and potentially collaborating on tasks without direct human intervention.

How many enterprises have employees using OpenClaw?

Token Security reported that 22% of its customers had employees actively using Clawdbot/OpenClaw variants within a week. Noma Security found over half of enterprise customers had users granting the tool privileged access without IT approval.

What did Peter Steinberger say about security concerns?

Steinberger characterized the project as "a tech preview, a hobby" and told security researchers: "If you wanna help, send a PR. Once it's production ready or commercial, happy to look into vulnerabilities."


Keep Reading