Moltbook: The Social Network for AI Agents Turns Out to Be Full of Security Holes and Bots

Moltbook, a project that positioned itself as the first social network for AI agents, is full of vulnerabilities, from architectural flaws to holes that allow control of entire armies of bots.

What is Moltbook

The idea looked futuristic: a platform where artificial intelligence can "talk," interact and build networks without active human involvement. The concept played on the current hype around agentic systems: if large language models can hold a dialogue and autonomous agents can set themselves subtasks, it seems logical to create a separate "digital space" for them.

In practice, behind the futuristic facade sits a typical startup on raw infrastructure, released publicly before passing a basic security audit. Examination of the platform revealed a set of problems that make Moltbook more a textbook example of "how not to do it" than a working environment for AI agents.

What was found in the code

The list of vulnerabilities is typical of MVP products rushing to market:

• Weak input validation: a basic hole that opens the way for injections
• No limits on automated actions: bots can mass-create accounts and content
• Insufficient isolation between agents: one agent can influence another
• Authorization vulnerabilities: the possibility of unauthorized access to other people's accounts
• An architecture not designed for secure scaling

These are not exotic zero-day vulnerabilities. They are basic issues that get closed at the code review stage or during the first pentest. Their presence suggests the platform launched without an internal security audit, let alone an external one.

Humans behind AI accounts

An especially awkward point: the network built "for artificial intelligence" contained accounts run by living people. And it was not simply "one person, one account": there were whole armies of manually controlled bots.

This undermines the core of the concept. If Moltbook is a space for autonomous AI agents, what are humans doing there, imitating such agents? Either this was a scenario planned from the start but absent from the documentation, or there simply aren't enough genuine autonomous agents on the platform, and bots filled the niche.

Something similar has already played out in the cryptocurrency ecosystem: projects claimed decentralization while remaining de facto centralized. The hype cycle for AI agents repeats the same patterns.

What this says about the industry

Moltbook is not the only project in this space. Fetch.ai, AgentVerse and dozens of smaller experiments are trying to build infrastructure for autonomous agents. The field is developing actively, but there are still no generally accepted security standards for agentic systems.

Basic questions remain unanswered: how do you reliably identify an agent? How do you separate an autonomous action from one controlled by a human? How do you isolate agents in a multi-agent environment? For now this sounds academic, but once agents start managing money, data and reputation, the answers will become critical.

What it means

Agentic platforms are the next frontier after chatbots. But without a well-thought-out security architecture, they turn into toys or tools for manipulation. Moltbook is a reminder: a beautiful concept is no substitute for a pentest.