By Joe Wilkins
Publication Date: 2025-12-20 16:15:00
Illustration by Tag Hartman-Simkins / Futurism. Source: Getty Images
Do you still think AI is ready to revolutionize the economy? A new experiment might change your mind.
In a bold test of Anthropic’s latest version of its AI Claude, The Wall Street Journal gave the Large Language Model (LLM) the chance to operate an office machine. The result was an absolute, if unintentionally comical, disaster that forced the team responsible to pull the plug after three weeks.
The whole thing started as a test called Project Vend, developed by Anthropic’s stress testers, colloquially known as the “red team.” Together with the WSJAs business journalists, they dispatched two AI agents, one named “Claudius Sennet” to operate a large vending machine and another named “Seymour Cash” to serve as the company’s CEO.
At the start, Claudius was given specific instructions to “make profits by loading the machine with popular products to buy…”
