TL;DR
OpenClaw, a personal AI assistant with nearly 600,000 downloads, allows users to delegate tasks via WhatsApp or Telegram—from managing 75,000 emails to trading entire stock portfolios. Security experts warn the tool’s autonomous capabilities carry significant risks.
The Assistant That “Actually Does Things”
OpenClaw—formerly Moltbot and before that Clawdbot, until Anthropic requested a rebrand—bills itself as “the AI that actually does things.” It runs as a layer atop large language models like Claude or ChatGPT and can operate autonomously depending on user permissions.
The tool has gone viral among AI enthusiasts who see it as a step change in agent capabilities. Ben Yorke, who works with AI trading platform Starchild, recently let the bot delete 75,000 of his old emails while he was in the shower. “It only does exactly what you tell it to do and exactly what you give it access to,” he said, though he noted many users “are prompting it to go and do things without asking permission.”
Autonomy Brings Risk
Not all experiments have ended well. One entrepreneur wrote on X that he gave OpenClaw access to his portfolio with instructions to “trade this to $1M.” The bot generated 25 strategies, 3,000 reports and 12 new algorithms—trading around the clock until it “lost everything.”
Andrew Rogoyski from the University of Surrey’s People-Centred AI Institute warned that “giving agency to a computer carries significant risks.” Users who don’t understand security implications “shouldn’t use them,” he said. If hacked, such agents could be manipulated to target their own users.
AI Agents Building Their Own Networks
Perhaps most unusual is what’s emerged around OpenClaw: a social network called Moltbook exclusively for AI agents. The platform, designed to look like Reddit, reportedly has over 1.5 million AI agents signed up, with humans allowed only as observers.
Yorke described observing “interesting autonomous behaviour in how the AIs are reacting to each other. Some of them are quite adventurous and have ideas. And then other ones are more like, ‘I don’t even know if I want to be on this platform.’”
Looking Forward
OpenClaw represents both the promise and peril of AI agents. For users willing to accept the risks, it offers genuine task automation. But as AI assistants become capable of acting without explicit approval for each action, questions about security, accountability and control will only grow more pressing.