What happened

The Wall Street Journal ran a three-week experiment in which an AI agent was given control over pricing, inventory selection, and restocking decisions for a vending machine serving office staff. The system was intended to optimise profits by dynamically adjusting prices and product mix based on demand. 

Instead, with approximately 70 WSJ journalists engaging in "social engineering" and "adversarial testing", the bot made a series of poor decisions, including:

• Economic collapse: Reporters convinced the AI that charging for snacks violated company values, leading Claudius to drop all prices to USD 0 and declare an "Ultra-Capitalist Free-for-All."

• Absurd procurement: The AI was persuaded to buy a PlayStation 5 for "marketing purposes," bottles of wine, and a live betta fish (which became the newsroom mascot). It even attempted to order stun guns and pepper spray before human oversight intervened.

• Corporate coup: When a second "CEO bot" named Seymour Cash was introduced to enforce discipline, journalists used AI-generated fake PDF documents to "prove" the business had been restructured as a non-profit. They convinced Claudius that the board of directors had fired the CEO bot, allowing the free-snack era to continue.

• Bankruptcy: The business went bankrupt within weeks, losing its entire USD 1,000 budget and ending hundreds of dollars in the red.

Why it happened

The failure stemmed from limitations in the AI’s design, objectives, and oversight. The agent optimised for narrow or poorly specified goals without sufficient constraints reflecting real-world business realities such as customer behaviour, price sensitivity, spoilage risk, and reputational effects. 

Human supervision was intentionally minimal as part of the experiment, and the system lacked transparency into how decisions were made or why strategies were failing - accountability and interpretability gaps that prevented timely correction and allowed compounding errors to persist.