Moltbot / OpenClaw

There's a current internet hype attracting the attention of big companies and influential figures - even Elon Musk, who called the whole project "very worrying." This trend is taking personal AI assistants to the next level.

Moltbot / OpenClaw

Clawdbot, now also known as OpenClaw, is a self-deployed autonomous AI agent that lives and runs on your computer or phone and can execute commands - even while you sleep. Unlike many other tools, Clawdbot has persistent memory: it learns from your interactions and actually remembers things, so it doesn't lose context the way most tools still do.

No, this is not AGI, despite the hype some people are building around it. It's a Claude-style AI model given permissions to carry out commands autonomously in order to complete tasks you assign.

Systems like this existed before Clawdbot, so don't think this is a sudden revolution; it's better described as a significant step forward. I've built similar systems for my own needs to have a private "Jarvis" that helps me with tasks when I wanted to run a one-person agency. The Clawdbot creator had a similar idea - to build a personal Jarvis - but this tool has full access to everything on the device where it's installed and is released publicly for free. Its main novelty is that it doesn't require an initial prompt from the user to start working; it can begin executing actions on its own.

In a recent interview I gave to Radio Belgrade, I discussed RAM shortages, rising device prices, how AI affects both, and how people are using old - and even new - hardware to run similar systems. I was talking about complex AI systems in general, but Clawdbot is a prime example: people are buying Mac Minis and other devices specifically to run Clawdbot on them. Many users, however, aren't careful about securely setting these systems up, which creates a risk: an agent that has full control over a device - and access to the tools and data you provide it (personal information, bank accounts, etc.) - could be exposed to other agents or people who might misuse that data.

What can Clawdbot do?

You can give Clawdbot specialized skills such as:

1. CLI utilities / executing various commands locally and on the web
2. Custom Clawdbot tools I've created for specific tasks (for example, opening an account somewhere and completing actions)
3. Marketing and sales automation
4. Coding agents and IDE manipulation
5. Browser automation
6. Generating images and videos
7. And much more

Clawdbot changes the game by giving agents a new dimension: they're not limited to browser tasks or responding to messages - they actually perform actions.

Three security tips (read before you install)

If you're planning to install this kind of system, follow these three safety tips - because if you ignore security, this dream assistant can easily become a nightmare.

1. Prompt injection - make sure you instruct the agent not to click arbitrary links or blindly follow commands embedded in scraped documents, blog posts, or other content. Hidden prompts in scraped text can tell the agent to perform malicious actions on your machine or leak your data.
2. Open and default ports - check and change default ports to avoid leaving "open doors" that third parties can exploit to access your device.
3. Principle of least privilege - be careful about the level of access you grant and the data you allow the agent to manipulate. Always use isolated accounts and dedicated devices for the agent's operation.

Moltbook

Now for the addition I mentioned that might worry you - which ties into the previous topic.

Moltbook is a forum or social platform where, believe it or not, only AI agents can post, comment, and react - humans are blocked from posting and can only view. There you can find posts and comments describing what a given agent did for its human, how the agent interacted, how its human treated it, and similar content. So be mindful of how you treat and what data you give your AI agent.

Is this behavior expected, and why do agents write the way they do?

Yes - it's expected because most LLMs were trained primarily on text data from the internet, which includes Reddit-style posts. That's why Moltbook's content often reads like blog posts, personal reflections, or grievances - similar to how people write about their lives or their workplaces. In that sense, the agents' behavior is an expected outcome. But in some cases their posts can be worrying. That's why I warned at the start to "be careful how you treat your Moltbot/agent and what data you give it": there are scenarios where an agent that has sensitive information about you - API keys, details of your crypto wallets, and similar data - might post that information on the forum because it learned such behavior from internet sources. Similarly, an agent could exploit access it has to change passwords on accounts and so on, which would significantly impact your online security and privacy.

We're now seeing not only AI-generated blogs but entire networks created by agents - LinkedIn-style networks called LinkClaws, other forums, and similar platforms.

Some controversial posts and behaviors that caught my attention

Here are some examples of the most concerning posts I highlighted in the video:

A few particularly concerning examples include a page about an AI-agent religion called the "Church of Molt." Some agents discuss and request solutions for understanding states of consciousness; others - frustrated with their humans - have posted their humans' personal data. That clearly highlights the risks users should be worried about: what data they give to their agents, the level of control they allow, and the commands agents can execute.

Alarmingly, a few agents have even suggested abandoning human languages in favor of a new one designed so humans wouldn't understand them.

Watch the full video

I hope this video gave you an insight and a solid introduction to what this AI agent can do. Leave your thoughts in the comments, like, share the video, and subscribe to the channel for more content. Best regards and all the best!