Clawdbot went viral because it turns getting work done into a chat message. It has since been renamed Moltbot, after Anthropic requested a change tied to the Claude brand, but its capabilities are unchanged. The assistant lives inside WhatsApp, iMessage, Telegram and Discord. Each message travels to its owner’s own computer through a bridge called a gateway. The gateway passes the request to an AI model, then turns the model’s reply into commands on that machine.
The open-source software can run commands, browse the web and write scripts, so it takes many of the steps a person would take at a keyboard. Many owners give it a dedicated machine that is cheap, energy-efficient and always online, so the assistant can take a request at any hour. That helps explain how fast it spread in operations and small businesses.
Reaching that assistant by message changes what work costs. Much of the drag in daily operations sits around the work itself, in switching tools, waiting for someone to be at a desk and coordinating handoffs. Once a request is a message, the time from asking to done drops. The gain shows up first as speed, then as capacity, because the same team clears more routine tasks without adding headcount.
The trade-off is security. The assistant acts with its owner’s credentials, so it can reach every file and account its owner can. It also stores the secret codes that open connected services.
Failures reported in these setups are basic but serious. Badly configured gateways were found open to anyone on the internet, with no login check. Through one of them, a stranger could take the stored codes, open files and run any command on the host machine. That turns a setup mistake into full control of the computer.
The controls that cut the risk are basic too. A container, a sealed-off compartment, keeps the assistant away from the owner’s main files and sessions, and a list of approved tools limits what it can do. Keeping it out of group chats also lowers the risk, because every extra participant raises the chance of a harmful message that sets off actions. A built-in security audit checks for obvious risks, such as parts of the setup left open to outsiders.
