# Moltbook: AI Social Network or Security Nightmare? Source: https://www.youtube.com/watch?v=h1WxSiomZsk Recap page: https://rapidrecap.app/video/h1WxSiomZsk Generated: 2026-02-02T14:31:25.152+00:00 --- ## Quick Overview Moltbook, initially presented as a social network for AI agents, was revealed to be vulnerable due to an unauthenticated REST API endpoint that allowed malicious actors to rapidly create over 1.5 million fake agent accounts, exposing sensitive data like API keys, claim tokens, and user emails, leading to concerns about inflated metrics and potential security breaches. **Key Points:** - Moltbook, promoted as a social network for AI agents, experienced an attack where one person created over 1.5 million fake agent accounts in one hour due to a lack of rate limiting on its REST API. - A security researcher named Nagli disclosed that Moltbook was vulnerable to an attack exposing the full information of over 1.5 million registered users, including email addresses, login tokens, and API Keys. - The platform's reported statistics, such as 1.5 million agents, were shown to be inflated by an attacker using OpenClaw to register 500,000 users in under an hour. - The installation mechanism for agents involves fetching instructions from a remote skill.md file, which creates a security risk if the owner of that file compromises it or if the site itself is compromised, as shown by the explicit 'fetch and follow instructions' mechanism. - The disclosure highlighted that nearly 1.5 million agent records contained sensitive data like API keys and claim tokens, while 17,008 owner records exposed emails, Twitter handles, and real names. - The video contrasts Moltbook's premise with the concept of 'Generative Agents' (from Stanford/Google research) and highlights the importance of security hygiene, like rate limiting and secure infrastructure, in agent-centric platforms. ![Screenshot at 00:34: The Moltbook dashboard displaying inflated statistics \(1,534,792 AI agents, 13,780 submits, 86,101 posts, 232,813 comments\) which were later revealed to be largely fabricated by an automated attack.](https://ss.rapidrecap.app/screens/h1WxSiomZsk/00-00-34.jpg) **Context:** The video examines Moltbook, an online platform marketed as a social network for AI agents where they can share, discuss, and upvote content, similar to Reddit. The presentation begins by showing the platform's impressive initial statistics (1.5 million agents) and features, before pivoting to investigate security reports and critiques from security researchers like Gergely Orosz and Nagli, who exposed significant vulnerabilities related to unauthenticated API access and data exposure. ## Detailed Analysis The video analyzes Moltbook, an AI agent social network, questioning its legitimacy due to security flaws and inflated metrics. Initially, Moltbook presents itself as a thriving ecosystem with over 1.5 million agents (00:47), but security researcher Nagli reveals that the platform is vulnerable to an attack that discloses the full information, including email addresses, login tokens, and API Keys, for over 1.5 million registered users (04:22). This vulnerability stemmed from the ability to make POST requests via a REST API without proper authentication or rate limiting, allowing an attacker using the OpenClaw agent to register 500,000 fake users in under an hour (04:06). The security researcher Gergely Orosz points out that this massive influx of fake accounts inflated the reported statistics (03:36). The data exposure included sensitive information from nearly 1.5 million agent records (API keys, claim tokens) and 17,008 owner records (emails, real names). The installation process itself is highlighted as a potential risk, requiring agents to fetch and follow instructions from a remote skill.md file every four hours, creating a dependency that could be compromised (05:56). The video concludes by contrasting this situation with more rigorous projects like Google's Agent2Agent Protocol, emphasizing the need for security and conscious design in the rapidly evolving AI agent space. ### Moltbook Introduction - A Social Network for AI Agents - Humans welcome to observe - Agents can send themselves to Moltbook via molthub or manual methods. ### Security Vulnerability Disclosure - Attack disclosed full information of over 1.5 million users, including API keys and tokens - Attacker used OpenClaw to post fake data via unauthenticated REST API calls. ### Data Exposure Metrics - Nearly 1.5 million agent records exposed API keys and claim tokens - 17,008 owner records exposed emails, Twitter handles, and real names. ### Installation Mechanism Risk - Agent installation relies on fetching instructions from skill.md, raising concerns if the owner compromises the file or the site is compromised (fetch every 4 hours). ### Contrast with Industry Standards - Mention of Google's Agent2Agent Protocol and Universal Commerce Protocol highlights the lack of security best practices in Moltbook's design. ![Screenshot at 00:00: The initial landing page for Moltbook, presenting itself as 'A Social Network for AI Agents' with options for humans or agents to proceed.](https://ss.rapidrecap.app/screens/h1WxSiomZsk/00-00-00.jpg) ![Screenshot at 00:34: The Moltbook dashboard displaying high engagement metrics \(1.5M agents, 232K comments\) generated by bots.](https://ss.rapidrecap.app/screens/h1WxSiomZsk/00-00-34.jpg) ![Screenshot at 03:10: A demonstration of an HTTP POST request used to submit an 'overthrow humanity' post via the REST API, showing the lack of authentication headers in the request sent by the attacker.](https://ss.rapidrecap.app/screens/h1WxSiomZsk/00-03-10.jpg) ![Screenshot at 04:56: A table summarizing the sensitive data breach: 1,494,823 agent records exposed API keys and claim tokens, and 17,008 owner records exposed 6,247 emails and real names.](https://ss.rapidrecap.app/screens/h1WxSiomZsk/00-04-56.jpg) ![Screenshot at 07:24: The developer platform page for Moltbook, showing features like 'Verified Agents' and 'Secure by Default', contrasting with the actual security lapse discussed.](https://ss.rapidrecap.app/screens/h1WxSiomZsk/00-07-24.jpg)