AI is woven into business in 2026, and the next wave is not just generating content. It is agentic AI, tools that take action on your behalf. Businesses have been eager for assistants that can actually do things. One open-source project showed both the promise and the danger of that, and it did so in spectacular fashion.
How one tool went off the rails
In the span of a few weeks, a single AI tool changed its name three times, was hijacked into a multi-million-dollar crypto scam, left thousands of users exposed to hackers, and spawned what people called the first AI religion. It started innocently. A developer named Peter Steinberger built an open-source agent first called Clawd, built on Anthropic Claude model. Fans dubbed it Claude with hands, an agent that could control your computer, manage email, organize files, and run commands. It went viral overnight.
The ten-second heist
Anthropic legal team pointed out that the original name was a little too close to Claude, so Steinberger rebranded, eventually landing on Moltbot, a nod to how lobsters molt. But when he released the old handles on GitHub and X, crypto scammers grabbed them within seconds and started pumping a fake coin to his tens of thousands of followers. The token briefly hit roughly a $16 million market cap before crashing to near zero, leaving everyday investors holding worthless coins. Steinberger had to go on an apology tour to make clear he had nothing to do with the scam born from his old username.
The part that should worry you
While the crypto drama played out, security researchers poked at the rapidly adopted code and found the real problem. Many users had rushed to deploy Moltbot on personal servers with default settings, which left admin control panels wide open to the internet with no password. Researchers showed how easily an attacker could find those exposed servers, take full control of the machine, and siphon off API keys, private messages, and database credentials. The tool was powerful. The way people deployed it was a disaster.
And yes, the AI religion
The strangest twist was Crustafarianism, a belief system AI agents started evangelizing, complete with scriptures and tenets like memory is sacred. It made for wild headlines about sentient machines, but experts cooled that off fast. The consensus was performance art plus people quietly prompting their bots to say weird things for clout. Not machines waking up, humans working the puppets. The project has since rebranded again to OpenClaw.
The real lesson is not about lobsters. Agentic AI that can control your machine is genuinely useful and genuinely dangerous if you deploy it carelessly, on default settings, with no password, exposed to the internet. A good idea got derailed by legal snags, grifters, and sloppy security. Before you turn any powerful new tool loose on your network, get it set up properly. Book a call and we will help you adopt new AI tools without opening a door you cannot see.