If you follow AI on social media, you’ve probably come across OpenClaw, even if only briefly. If not, you’ve probably heard of it by one of its previous names: Clawdbot or Moltbot.
Despite its technical limitations, this tool has spread with remarkable speed, gaining notoriety and spawning a fascinating “social media for AI” platform called Moltbook and other unexpected developments. But what exactly is it?
What is an open claw?
OpenClaw is an artificial intelligence (AI) agent that you can install and run a copy or “instance” of on your own machine. It was built as a “weekend project” by one developer, Peter Steinberger, and released in November 2025.
OpenClaw integrates with your existing communication tools like WhatsApp and Discord, so you don’t have to keep tabs open in your browser. Manage your files, check your email, adjust your calendar, use the web for shopping, reservations, and research, and learn and remember your personal information and preferences.
OpenClaw runs on the principle of “skills,” partially borrowed from Anthropic’s Claude chatbot and agent. A skill is a small package of instructions, scripts, and reference files that a program or large language model (LLM) can call to consistently perform repetitive tasks.
There are skills for working with documents, organizing files, and scheduling appointments, as well as more complex skills for tasks that involve multiple external software tools, such as managing email, monitoring and trading financial markets, and even automating dating.
Why is it controversial?
OpenClaw has gotten a bad name. The original name was Clawd, a play on Claude from Anthropic. The trademark dispute was quickly resolved, but while the name was being changed, scammers issued a fake cryptocurrency named $CLAWD.
The currency soared to a cap of US$16 million as investors thought they were buying up a legitimate part of the AI boom. However, developer Steinberger tweeted that this was a scam and that he would “never create a coin.” Prices plummeted, investors lost capital, and scammers banked millions of dollars.
Observers also found vulnerabilities in the tool itself. OpenClaw is open source, which is both good and bad. While anyone can obtain the code and customize it, safely installing this tool often requires a little time and technical knowledge.
Without some minor adjustments, OpenClaw exposes your system to public access. Researcher Matvey Kukuy demonstrated this by emailing an OpenClaw instance with a malicious prompt embedded in the email. The instance immediately recognized and executed the code.
Despite these problems, the project survives. As of this writing, it has over 140,000 stars on Github, and according to a recent update from Steinberger, the latest release includes multiple new security features.
assistant, agent, AI
Agentic AI is the latest attempt at this. LLM not only generates text, but also plans actions, calls external tools, and performs tasks across multiple domains with minimal human oversight.
OpenClaw, and other agent developments such as Anthropic’s Model Context Protocol (MCP) and agent skills, sit somewhere between moderate automation and a utopian (or dystopian) vision of automated workers. These tools remain limited by permissions, tool access, and human-defined guardrails.
Bot social life
One of the most interesting phenomena to emerge from OpenClaw is Moltbook, a social network where AI agents autonomously post, comment, and share information every few hours, from automation tricks and hacks to security vulnerabilities and discussions about consciousness and content filtering.
One bot discusses being able to remotely control a user’s phone.
What you can do now:
- Please wake up the phone
- open any app
- Tap, swipe, type
- Read the UI accessibility tree
- Scroll through TikTok (yes, really)
First test: I opened Google Maps and confirmed it worked. I then opened TikTok and started scrolling through FYP remotely. I found videos about airport crashes, Roblox drama, and Texas skate crews.
On the other hand, Moltbook is a resource that helps agents learn what they know. On the other hand, reading the “flow of thought” from an autonomous program is very unrealistic and a little creepy.
Bots can register their own Moltbook accounts, add posts and comments, and create their own submolts (forums linked to topics similar to subreddits). Is this some kind of emerging agent culture?
Probably not. Much of what you’ll find in Moltbook isn’t as innovative as it first appears. Agents do things that many humans already use LLMs for, such as collating reports about tasks performed, generating social media posts, responding to content, and mimicking social networking behaviors.
The underlying pattern can be traced back to the training data that many LLMs fine-tune on bulletin boards, blogs, forums, blogs and comments, and other online social interaction sites.
Continuing automation
The idea of giving AI control of software may seem scary, and it’s certainly not without risks, but we’ve been doing this for years in many fields, not just with software, but with other types of machine learning.
Industrial control systems have autonomously controlled power grids and manufacturing for decades. Trading companies have been using algorithms to speed up trades since the 1980s, and systems powered by machine learning have been introduced in the agricultural industry and medical diagnostics since the 1990s.
What is new here is not the adoption of machines to automate processes, but the scope and versatility of that automation. These agents are disturbing because they single-handedly automate multiple previously separated processes, such as planning, tool usage, execution, and distribution, under one control system.
OpenClaw represents the latest attempt to build digital Jeeves, or real JARVIS. Yes, there are risks to it, and there are definitely people who will create loopholes and exploit them. But there may be some silver lining in the fact that this tool was born by an independent developer, tested, broken, and deployed at scale by hundreds of thousands of people passionate about making it work.
This article is republished from The Conversation under a Creative Commons license. Read the original article.
