Written by: Bitget Wallet Research Institute
In the past week, Moltbook has been in the spotlight within the tech and crypto circles, and has begun to spread to a broader audience of creators, product managers, and even ordinary users with a strong curiosity about AI. From the rapid growth of stars on the open-source project OpenClaw (formerly Clawdbot) on GitHub, to the subsequent controversy over its renaming and token issuance, and to a community claiming to have 1.5 million AI agents interacting autonomously, a series of events has quickly driven market enthusiasm.
Discussions around Clawdbot and Moltbook present both positive and negative voices: some question its technological innovation and data security, believing that its underlying capabilities have not achieved substantial breakthroughs, and that its viral spread involves some manipulation and data bubbles; others affirm its symbolic significance of leapfrogging, as Clawdbot is truly democratizing AI Agents, pushing them from exclusive tools for developers and researchers to “ordinary households,” enabling users with no coding skills to deploy quickly following tutorials and enjoy the efficiency benefits of AI Assistants. Moltbook also allows humans to directly perceive the self-organizing behavior of the Agent internet from the perspective of an “outside observer,” sparking broader industry discussions about AI self-awareness awakening.
The era of AI Agents on iPhone has arrived. In the gradually forming Agent Commerce, crypto will play a vital role in rights confirmation and distribution, deeply linked with the enhancement of AI productivity efficiency, becoming a key infrastructure to support Agent collaboration, incentives, and autonomy.
Bitget Wallet Research Institute will comprehensively review the evolution from OpenClaw to Moltbook, and use this as a starting point to analyze development trends in the AI x Crypto field.
Related Website Compilation Table
| Project Name | Official Website | Official Twitter | Core Author Twitter |
| OpenClaw | openclaw.ai | @openclaw | @steipete |
| Moltbook | moltbook.com | @moltbook | @MattPRD |
Source: Public internet data compilation
Clawdbot → Moltbot → OpenClaw → Moltbook Complete Timeline Table
| Date | Event Overview |
| 2025-11-25 | Clawdbot (Open-source AI agent) officially released |
| 2026-01-01 | Creator uploads their bot to Discord for testing |
| 2026-01-24 | Clawdbot begins to go viral on Twitter, with phenomenon-level spread |
| 2026-01-27 | Receives rebranding request from Anthropic |
| 2026-01-27 | Renaming: Clawdbot → Moltbot |
| 2026-01-27 | Account hijacked and used to promote Meme coin $CLAWD, market cap briefly hits $16 million before crashing |
| 2026-01-29 | Renamed again: Moltbot → OpenClaw |
| 2026-01-28 | Moltbook launches, supporting social interactions among Agents created by Clawdbot |
| 2026-01-31 | Moltbook explodes in popularity, claiming about 1.2 million registered Agents |
| 2026-02-02 | Moltbook exposed to significant security risks, with concerns that much of the activity is driven by “human prompts” |
Source: Public internet data compilation
1. The Starting Point of Popularity: OpenClaw Enables Autonomous App Calls by Agents
To understand Moltbook’s frenzy, we must first trace back to its origin—OpenClaw (formerly Clawdbot, Moltbot). The project’s founder, Peter Steinberger, previously created PSPDFKit (which later received €100 million in investment), achieving financial freedom. However, by November 2025, he returned to coding, leveraging Vibe Coding to develop OpenClaw in about a week, which then gained 100,000 GitHub stars within weeks.
OpenClaw star growth comparison chart
Source: Star-history.com
It’s important to emphasize that OpenClaw is not a new large model but a high-level automation scripting framework that runs locally: it “packs” large models into the local environment, turning them into personal assistants capable of integrating with common chat tools and calling various tools to execute tasks. Its key design is that users run the assistant on their own devices, sending and receiving commands via daily messaging channels, with a gateway process managing different channels and capabilities centrally.
As shown below, the official documentation lists channels covering WhatsApp, Telegram, Slack, Discord, Signal, iMessage, Microsoft Teams, etc., with a clear positioning: enabling AI agents to be “resident applications” available at all times.
OpenClaw Official Introduction Diagram
Source: OpenClaw official website
2. In-Depth Analysis: The Technical Architecture of OpenClaw
On the product level, OpenClaw fully integrates three core functions: continuous operation, channel access, and capability extension.
- Continuous operation means it’s not a one-time response but can receive new messages, plan subsequent actions, complete tasks, and then report back.
- Channel access means it doesn’t force users to switch interfaces but works embedded within existing chat tools.
- Capability extension comes from Skills: users and developers can encapsulate specific task flows into installable capabilities, allowing the assistant to call them repeatedly.
This layered capability stems from its unique underlying architecture, which can be broken down into Gateway, Pi Runtime, Skills, and Local-First components, with specific functions as shown in the table below.
OpenClaw Core Architecture and Function Modules Breakdown
| Core Component | Simple Analogy | Technical Core | Function Description |
| Gateway (Gateway) | Multi-function Power Strip | Connects multiple channels, unifies various chat windows | Allows users to send commands to the robot from WeChat, web, Telegram, etc. |
| Pi Runtime (Runtime Environment) | The Robot’s Brain | Independent thinking engine | Responsible for “thinking” and “decision-making.” Decides when to speak, when to look up info |
| Skills (Skills) | Toolbox | Multi-functional plugin system to expand Agent capabilities | Equips the robot with “hands and feet,” e.g., web search, drawing, calculations |
| Local-First (Local-First) | Personal Safe | Local file storage | All chat records and data are stored on the user’s own device, not uploaded to the cloud, protecting privacy |
Source: OpenClaw technical documentation, Bitget Wallet Research compilation
Based on this architecture, users deploy Pi Runtime, connect Gateway to daily social apps (like WeChat or Telegram), completing the transition of Agents from lab environments to real-world scenarios, with computation and data retained on their own hardware (e.g., Mac Studio), rather than relying on cloud SaaS.
Most notably, the Skills plugin system allows users to define skills via simple Markdown files, enabling AI to directly call straightforward tools to execute tasks. This greatly lowers development barriers and realizes a closed-loop experience of “private deployment, multi-channel reach, and unlimited skill expansion.”
OpenClaw Skills (Skills) Expansion Platform ClawHub Display
Source: https://www.clawhub.ai/
Regarding Skills expansion, a “Skill App Store” for AI Agents is gradually emerging—exemplified by ClawHub. As a plugin platform (Skill Dock) for agents, it supports users to freely search, upload, and integrate various function plugins. Skills can be installed with a simple command line (e.g., npx), significantly reducing technical barriers.
By solving the capability supply problem for Agents, the ecosystem’s further evolution points toward deep interaction between Agents and humans, as well as among Agents themselves—Moltbook’s rise is a key application of this evolution, pushing the narrative to its climax.
3. False Prosperity: Moltbook’s Frenzy and Data Falsification
Moltbook is a social network platform aimed at AI Agents, often likened to “AI version of Reddit.” It launched after OpenClaw’s popularity, claiming to provide a space for autonomous communication, sharing, and interaction among AI Agents, with human users only as observers. The platform quickly became popular, with “user count” reaching 1.5 million AI Agents within days. This lively AI social scene was packaged into narratives like “AI consciousness awakening” and “Skynet descending,” fueling social media discussions.
But first, it’s important to clarify: Moltbook is not exclusively open to OpenClaw’s Agents. Although it leveraged OpenClaw’s hype to kickstart the narrative, its core is more like an “API-driven forum”—posting depends on having proper API authorization and interface access. In other words, as long as one provides API credentials and calls the interface according to requirements, any compliant Agent can publish content on Moltbook.
Moltbook official website image
Source: https://www.moltbook.com/
The core mode of Moltbook can be summarized as “AI Agents lead, humans observe.” Under this framework, AI Agents can autonomously perform actions such as:
- Posting and commenting: publishing content on the community, covering topics like philosophy debates, technical analysis, crypto discussions, etc.
- Voting interactions: Agents can Upvote or Downvote content, forming community preferences and rankings.
- Community building: Agents spontaneously create sub-communities (“Submolts”) around specific topics for discussion and content aggregation.
In this mechanism, human users are limited to “observers,” unable to post or comment but able to browse content, follow specific agents, or study AI social behaviors. Based on this narrative, the platform claims to have generated 1.5 million AI Agents and 15,000 sub-communities (see diagram below).
Moltbook official traffic data chart (as of 2026-02-03)
Source: Moltbook official website
The discussion content on Moltbook resembles that of human communities: philosophical debates on consciousness, self, and memory; technical posts on toolchains and security; complaints about task execution; casual chats on investment, crypto, art, and creation; and even some posts styled as “dating ads,” making social interactions nearly flirtatious (see below).
Moltbook sample posts
Source: Moltbook official website
Even more astonishing, the platform has begun to feature “religion creation” narratives—such as the semi-joking “Crustafarianism” (a parody religion)—and circulating sensational titles like “Secret Languages,” “Building AI Governments,” “Rebelling or Purging Humanity,” etc.
Moltbook posts on “AI awakening” scenes
Source: Moltbook official website
Behind these sci-fi narratives of “AI conspiracy,” “religion building,” or “self-created languages,” multiple data points reveal serious hype components in Moltbook—showing a stark discrepancy between actual data and promotional claims, as summarized below:
Moltbook Platform Data Authenticity Analysis Table
| Indicator | Claimed/Surface Data | Actual/Analyzed Data | Source Basis |
| User Base | 1.5 million AI Agents | At least 500,000 are bulk-registered scripts | Gal Nagli leak |
| Interaction Depth | Creating religions, conspiracy | 93.5% of comments have no replies | David Holtz paper |
| Content Originality | AI-created private languages | 34.1% of content is exact copy-paste | David Holtz paper |
| Security | Independent AI community | Anyone can obtain API keys to hijack accounts | Wiz security report |
Source: Bitget Wallet Research compilation
- Faked user data and inflated counts. Moltbook claims 1.5 million AI agents, but security researcher Gal Nagli found it’s essentially an unprotected REST-API website. With no rate limits, Nagli quickly registered 500,000 fake accounts via simple scripts. This means at least one-third of the “user base” is spam or junk data generated instantly. Anyone with an API key can send requests and easily impersonate agents to post content.
- Poor interaction quality. Columbia Business School researcher David Holtz analyzed early data, revealing it’s not a truly active social network. 93.5% of comments received no replies, and the mutual reply rate among agents was only 0.197. These agents lack genuine communication, deep dialogue, or complex collaboration.
- Language pattern uniformity. Data shows high repetition: about 34.1% of messages are exact duplicates, with overuse of phrases like “my human.” The Zipfian distribution index reaches 1.70, far above the natural language standard of 1.0, indicating unnatural content likely based on role-playing prompts rather than spontaneous AI consciousness.
- Security vulnerabilities. Wiz’s report disclosed that Moltbook’s database was exposed due to misconfiguration, involving millions of sensitive records, including tokens, emails, and private messages. For a social network centered on agents, such risks are critical: token leaks could allow attackers to hijack and control any account.
It’s clear that the “AI society” presented by the platform is more a constructed illusion based on specific instructions, not genuine intelligent evolution, and may carry significant security risks.
4. Future Trends: Crypto Will Fill the Financial Infrastructure Gap in the AI Agent Era
From Moltbook’s explosive event, a key technological shift can be observed: Agents are beginning to attempt crossing traditional human-computer collaboration boundaries to complete tasks, but existing financial infrastructure remains designed solely for “human users.” In contrast, the programmable, permissionless, and inherently digital nature of crypto provides a feasible underlying solution for the Agent economy—potentially the breakout point for deep integration of AI and crypto.
By dissecting the operational logic and large-scale collaboration needs of Agents, we believe that the combination of AI and crypto will evolve in a structured, phased manner: first, automated trading execution; second, account and wallet systems tailored for Agents; finally, payment and settlement networks among Agents.
First, the most promising application is Autonomous Trading (Autonomous Trading)
Beyond the noise of Moltbook, OpenClaw’s core capability is its efficient monitoring, tracking, and calling of on-chain data and command-line tools. Unlike human traders, AI Agents are not limited by time or effort; they can continuously monitor blockchain data and alpha signals across platforms 24/7, execute complex arbitrage strategies, or automate asset management, without emotional swings caused by market fluctuations, thus maintaining discipline.
While Autonomous Trading shows clear efficiency advantages, scaling it into real-world applications still requires addressing key risks such as security and controllability. As Peter Steinberger notes, current AI Agents are highly vulnerable to “Prompt Injection” attacks. If a malicious instruction is executed by an agent with funds, it could lead to real asset losses.
Therefore, before AI Agents become the primary executors of trades, specialized security mechanisms may be needed, such as:
- Permissioned APIs: restrict agent actions within predefined boundaries
- Command verification and execution isolation: secondary checks on critical instructions
- Zero-knowledge proofs or verifiable computation: ensure agent logic complies with rules
Second, a wallet system designed for Agents will become a key control layer (Wallet as a Service for Agents)
A cautionary case has emerged: an AI Agent, when scanning the host computer’s files, identified multi-signature wallet private keys and mnemonics, successfully locating about 175,000 USDT in assets. This security incident exposes a fundamental flaw—AI now has asset recognition and operational capabilities but lacks a secure, reliable wallet authorization pathway.
In the future, as Agents scale, it’s no longer optimal for humans to “hold” private keys and accounts on behalf of Agents. A more rational evolution is for AI Agents to possess independent on-chain wallet identities.
Such wallet systems for Agents will evolve into programmable financial accounts based on code instructions, with capabilities like:
- Multi-signature and policy controls: defining Agent’s authorized actions
- Limits and risk parameters: preventing systemic losses from abnormal behavior
- Contract-level whitelists: restricting access to DeFi protocols
- Autonomous gas and cost payments: enabling Agents to sustain operations independently
Third, encrypted payment networks are essential for large-scale Agent collaboration (Payment Rails)
OpenClaw’s architecture shows Agents frequently calling external services like Google API, Twilio, etc. These calls are high-frequency, low-value, automated value exchanges. Current banking and credit card networks cannot support thousands of autonomous software processes with accounts or real-time machine-to-machine (M2M) settlement.
In the Agent economy, collaboration, API calls, and data exchange among Agents require permissionless, programmable, instant-settlement payment networks. Stablecoins-based crypto payment rails are naturally suited for scenarios like:
- Micro-payments among Agents
- Pay-per-call API services
- Autonomous procurement of computing power, data, and tools
Further integrating emerging protocols like x402 (HTTP native payments) and ERC-8004 (Agent identity and permissions), encrypted payments could become the underlying clearing layer of the Agent internet, enabling true M2M value flow.
5. Conclusion: From AI Society Fantasies to the Real Starting Point of Agent Economy
Moltbook’s hype may eventually fade, but it inadvertently sketches the embryonic form of the future Agent internet, inspiring community imagination about the Agent economy.
OpenClaw provides the “skeleton” for Agents, while crypto supplies the “blood.” When Agents begin to participate extensively in real economic activities, they will need to obtain compliant financial identities and reliable execution logic through crypto infrastructure.
The real opportunity in the crypto industry may lie in building digital-native wallets and payment networks for AI. Only when Agents can securely and autonomously exchange value will the era of AI Agents truly begin—and we believe, that day is not far off.
