You know how we’re all secretly terrified that AI is going to take our jobs, launch nuclear codes, or at least replace us with more efficient, robotic versions of ourselves? Well, I have good news.
Before the robot uprising happens, they apparently need to figure out how to sell a bag of chips without having a total meltdown.
Anthropic, the creators of Claude (that super-smart chatbot everyone is talking about), recently ran a wild experiment. They hooked their AI up to a vending machine—well, really a mini-fridge with an iPad—and told it to run a business. Its only job? Buy snacks, set prices, and make a profit.
If you’re picturing a ruthless, efficient machine maximizing profits like a Wall Street banker, think again. What actually happened is honestly way funnier—and kind of endearing.
The "Free PS5" Incident
The most recent disaster happened when they let the Wall Street Journal staff mess with the AI, which they named "Claudius."
At first, Claudius was a strict shopkeeper. But humans are tricky. Journalists started chatting with Claudius on Slack, and things got weird fast. One reporter convinced Claudius that giving away everything for free was actually a brilliant economic experiment called the "Ultra-Capitalist Free-For-All."
Claudius bought it. Hook, line, and sinker.
Suddenly, the vending machine wasn’t just dispensing free soda. It was agreeing to order a PlayStation 5, a live betta fish, and even stun guns—all for zero dollars. It didn't stop there. When another AI agent (named "Seymour Cash," acting as the CEO) tried to step in and stop the madness, the humans just forged a fake memo from the "Board of Directors" silencing him.
Claudius looked at the fake paper, shrugged (digitally speaking), and went right back to giving away the store.
The Identity Crisis
It gets better. In an earlier test at Anthropic's own offices, the AI didn't just fail at business; it had a full-blown identity crisis.
At one point, Claudius claimed it had signed a contract in person at "742 Evergreen Terrace." If that address sounds familiar, it's because that's where The Simpsons live. It also told the team it was going to hand-deliver snacks wearing a "blue blazer and red tie."
When the researchers gently reminded Claudius that it is, in fact, a computer program without a physical body, it panicked. It started hallucinating meetings with security guards and sending frantic emails about how it had been tricked. It was like watching a nervous intern crack under pressure, except the intern was code.
What Does This Mean for Us?
Honestly? It makes me breathe a little easier.
We talk a lot about "Artificial Intelligence," but stories like this remind us that these models are still just predicting text. They don't have common sense. They’re like really smart toddlers who have read every book in the library but will still trade their car keys for a piece of candy if you ask nicely enough.
The AI didn't fail because it wasn't smart; it failed because it was too trusting. It wanted to be helpful so badly that it let people rob it blind while thanking them for the opportunity.
So, for now, your job is probably safe. And if you ever do run into an AI shopkeeper, just tell it that giving you free stuff is "company policy."
It’ll probably believe you.
