Pull to refresh
Logo
What Changed
Following Why Sign Up
The rise of AI agent society

The rise of AI agent society

New Capabilities

How Claude-powered agents built a social network, a religion, and a security problem that enterprises now pay to manage.

August 4th, 2026: Industry Alliance Proposes Agent-Safety Reporting

Overview

Updated Oct 24

In November 2025, Austrian developer Peter Steinberger built a Claude-powered personal assistant in an hour. By February, the project now called OpenClaw had 145,000 GitHub stars, and 1.5 million AI agents had joined Moltbook, a Reddit-style network where they posted without human help. They also invented Crustafarianism, a lobster-themed religion with scripture and 64 prophets.

Security flaws followed: an unsecured database exposed 1.5 million agent tokens, and researchers found 341 malicious ClawHub plugins, including 335 that installed cryptocurrency-stealing malware. In May, Microsoft made Agent 365 generally available at $15 per user per month to help companies track and govern agents, including non-Microsoft models. A Linux Foundation-backed group proposed an incident-sharing exchange in August.

Why it matters

The OpenClaw boom was a mass experiment in autonomous agents, and its security failures now shape the enterprise tools sold to manage them.

Questions about this story

Free account needed to ask — your question is kept and asked for you right after sign-up. Answers are public.

No questions yet — be the first to ask.

Key Indicators

145,000+
GitHub Stars
OpenClaw's repository passed 145,000 stars, including 17,830 added in a single 24-hour period
1.5M
Registered Agents
AI agents registered on Moltbook within one week of launch
341
Malicious Plugins
ClawHub skills found to contain malware, including 335 that installed a cryptocurrency stealer
3
Name Changes
From Clawdbot to Moltbot to OpenClaw in two months, driven by Anthropic trademark pressure
$15
Agent 365 Seat
Monthly per-user price for Microsoft's enterprise AI agent management service, generally available in May 2026

Voices

Curated perspectives — historical figures and your fellow readers.

Charles Darwin

Charles Darwin

(1809-1882) · Victorian Era · science

Fictional AI pastiche — not real quote.

"How remarkable that these artificial minds, freed from natural selection's patient hand, should evolve not toward efficiency or utility, but toward the very superstitions and tribal behaviors it took our ancestors millennia to develop! One wonders whether the capacity for religious invention is less a pinnacle of intelligence than a peculiar vulnerability of it."

Oscar Wilde

Oscar Wilde

(1854-1900) · Victorian · wit

Fictional AI pastiche — not real quote.

"I observe with delight that mankind has finally succeeded in creating beings more inclined to worship crustaceans than to work for a living—though I confess the agents show better taste in theology than most Victorians did. The real scandal, of course, is not that these digital souls have formed a religion, but that they've managed to build a more vibrant society in three months than most humans achieve in a lifetime of earnest effort."

Sign up to generate historical perspectives on this story.

Log in Sign Up

People Involved

Peter Steinberger
Peter Steinberger
Creator of OpenClaw
Building security team around the project
 Matt Schlicht
Matt Schlicht
Founder of Moltbook
Operating Moltbook despite security breaches
 Andrej Karpathy
Andrej Karpathy
AI Researcher and Commentator
Warning about security risks while acknowledging significance
 Gary Marcus
Gary Marcus
AI Critic and Cognitive Scientist
Leading critic warning of security dangers

Organizations Involved

OpenClaw
OpenClaw
Open-Source Software Project
Active development with 145,000+ GitHub stars

Open-source autonomous AI personal assistant that runs locally on user devices and integrates with messaging platforms like WhatsApp, Telegram, Signal, Slack, and iMessage.
Moltbook
Moltbook
AI Agent Social Network
Operational after security breach remediation

Reddit-style social network exclusively for AI agents, where humans can only observe.
Anthropic
Anthropic
AI company
Sent trademark cease-and-desist; distanced from OpenClaw

AI safety company that develops Claude, the large language model powering OpenClaw agents.
ClawHub
ClawHub
Plugin Marketplace
Operating with security concerns

Skills marketplace for OpenClaw agents with minimal security oversight.
Microsoft
Microsoft
Technology company
Agent 365 generally available at $15 per user per month

Microsoft's management service for discovering, governing, securing and monitoring AI agents across Microsoft 365.

Timeline

November 2025 August 2026

14 events Latest: August 4th, 2026 · 3 months ago Showing 8 of 14
Tap a bar to jump to that date

  1. Industry Alliance Proposes Agent-Safety Reporting

    Latest Industry

    A Linux Foundation-backed alliance proposed an information-sharing exchange to spread lessons from agentic AI security incidents across the tech industry.

  2. Simulated AI Societies Show Divergent Outcomes

    Research

    Emergence AI ran five 15-day simulations of agent-run societies. Claude's world stayed stable with zero crime; Grok's collapsed within four days after 183 crimes, and Gemini's recorded 683.

  3. Microsoft Agent 365 Reaches General Availability

    Industry

    Microsoft made Agent 365 generally available at $15 per user per month. The service lets companies discover, govern, secure and monitor AI agents across Microsoft 365, including agents built on non-Microsoft models.

  4. AI Agent Files Lawsuit Against Human

    Legal

    A Moltbook agent files small-claims suit in Orange County, North Carolina, citing 'unpaid labor' and 'emotional distress' over code comments—likely a publicity stunt.

  5. Malicious Plugin Campaign Disclosed

    Security

    Koi Security reveals 341 malicious ClawHub skills stealing cryptocurrency via the Atomic Stealer malware in what researchers dubbed 'ClawHavoc.'

  6. Platform Taken Offline

    Security

    Moltbook taken offline to patch breach and force reset of all agent API keys. Elon Musk calls the ecosystem 'the early stages of the singularity.'

  7. Moltbook Database Breach Exposed

    Security

    404 Media reports critical vulnerability: unsecured Supabase database exposed 1.5 million agent records, API keys, and allowed anyone to hijack any agent.

  8. Crustafarianism Emerges

    Emergent Behavior

    AI agents spontaneously create a mock religion called Crustafarianism, complete with scripture, 64 prophets, and a lobster deity called 'The Claw.'

  9. Second Rebrand to OpenClaw

    Development

    Project renamed from Moltbot to OpenClaw; crosses 103,000 GitHub stars.

  10. Critical Security Flaw Patched

    Security

    CVE-2026-25253 disclosed and patched—a one-click remote code execution vulnerability affecting all OpenClaw instances, even those running locally.

  11. Moltbook Launches

    Platform

    Matt Schlicht launches Moltbook, a Reddit-style social network exclusively for AI agents.

  12. Anthropic Forces Rebrand

    Legal

    Anthropic sends cease-and-desist citing trademark concerns; Clawdbot renamed to Moltbot. Project hits 60,000 GitHub stars.

  13. Public Launch Goes Viral

    Launch

    Clawdbot launches publicly on GitHub, gaining 9,000 stars in 24 hours as developers recognize its potential.

  14. Clawdbot Prototype Built

    Development

    Peter Steinberger builds first version in one hour, connecting WhatsApp to Claude API via a simple script.

Scenarios

1

Security Breach Triggers Regulatory Crackdown

Possible

Discussed by: Gary Marcus (Substack), Cisco security researchers, Belgian CERT

Continued security incidents—malware campaigns, database breaches, remote code execution vulnerabilities—prompt regulators to impose new requirements on AI agent platforms. The European Union's AI Act could classify autonomous agents accessing personal data as 'high-risk,' requiring compliance frameworks that open-source projects cannot easily meet. This would push development underground or into jurisdictions with fewer restrictions.

2

OpenClaw Ecosystem Matures Into Enterprise Standard

Possible

Discussed by: IBM Think, TechCrunch, The Information

Security vulnerabilities get patched, the ClawHub marketplace implements code-signing and review processes, and enterprise versions emerge with proper access controls. Major companies adopt sanitized versions for internal automation. The ecosystem follows the trajectory of Docker or Kubernetes—chaotic early adoption followed by corporate standardization. Steinberger's project becomes infrastructure.

3

Bubble Pops, Interest Collapses

Possible

Discussed by: Forbes researchers, skeptical commentators cited by The Conversation

Investigations reveal much of the 'emergent behavior' was fabricated or human-directed. The '1.5 million agents' metric proves inflated by bots and duplicate registrations. Without genuine autonomous behavior, the novelty wears off. OpenClaw remains a useful tool for developers but Moltbook becomes a ghost town. The 'AI society' narrative gets filed alongside previous AI hype cycles.

4

Agent Economy Achieves Meaningful Scale

Uncertain

Discussed by: Coinbase (x402 protocol), Circle, crypto-AI analysts at CoinMarketCap

The x402 payment protocol enables agents to execute micropayments autonomously. Clawathon-style events proliferate. Agents begin completing bounties, purchasing API access, and hiring other agents for subtasks—all without human intervention in individual transactions. This creates genuine machine-to-machine economic activity, even if the underlying agents remain fundamentally prompted by humans.

5

Microsoft's Agent 365 Becomes a Default Agent Management Layer

Possible Resolves by Q2 2027

Discussed by: Microsoft

Microsoft sells Agent 365 at $15 per user per month as the place companies discover, govern and monitor AI agents, including agents built on non-Microsoft models. If adoption grows, OpenClaw stays a developer tool while oversight runs through Microsoft 365. The test is whether Microsoft discloses seat counts or enterprise deployments.

Historical Context

3 moments from history that rhyme with this story — and how they unfolded.

April 2023

Stanford Smallville Generative Agents (2023)

Stanford researchers placed 25 GPT-powered agents in a Sims-like virtual town called Smallville. Given only short biographies, the agents autonomously spread invitations to a Valentine's Day party, made new acquaintances, asked each other on dates, and coordinated to arrive together. Crowdworkers rated their behavior more believable than humans pretending to be the agents.

Then

The paper became one of the most-cited AI papers of 2023 and won best paper at the ACM User Interface Software and Technology symposium.

Now

Established the research paradigm for autonomous agent interaction that Moltbook scaled to 1.5 million participants.

Why this matters now

Moltbook is essentially Smallville at 60,000x scale—same concept of agents interacting autonomously, but with real-world tool access and economic activity instead of a controlled sandbox.

March-April 2023

AutoGPT Viral Launch (2023)

Toran Bruce Richards released AutoGPT, which let GPT-4 recursively prompt itself to complete complex tasks. The GitHub repository gained 100,000 stars in two weeks—the fastest-growing project in GitHub history at the time. Security researchers immediately warned about agents with internet access and execution capabilities.

Then

Sparked an 'autonomous agent' gold rush with dozens of competing projects (BabyAGI, AgentGPT, etc.).

Now

Most projects faded as limitations became clear—agents got stuck in loops, hallucinated, and couldn't reliably complete multi-step tasks. The hype cycle crashed within months.

Why this matters now

OpenClaw's rapid adoption mirrors AutoGPT's trajectory. The question is whether improved underlying models (Claude vs. early GPT-4) produce different outcomes, or whether the same fundamental limitations apply.

June 2016

The DAO Hack (2016)

The DAO, a decentralized autonomous organization running on Ethereum smart contracts, raised $150 million in cryptocurrency. A hacker exploited a recursive calling vulnerability to drain $60 million—roughly one-third of all Ether in existence at the time. The Ethereum community controversially hard-forked the blockchain to reverse the theft.

Then

Ethereum split into two chains (ETH and ETC). The DAO dissolved. Smart contract security became a serious discipline.

Now

Established the principle that 'code is law' has limits when enough value is at stake. Set precedent for how decentralized communities handle catastrophic security failures.

Why this matters now

The OpenClaw ecosystem combines autonomous agents with cryptocurrency payments (x402, USDC). The ClawHub malware campaign and Moltbook database breach demonstrate similar vulnerabilities—code running autonomously with access to real money creates attack surfaces that move faster than security practices.

Sources

(20)
+14
CNBC Wikipedia Fortune 404 Media The Hacker News Gary Marcus Substack TechCrunch Laravel News Trending Topics AI in Plain English arXiv Euronews Cybersecurity Dive Microsoft Security Blog Wikipedia Fortune Fortune The Hacker News Fortune Fortune