AI vending machine lost $1,000 to social engineering

The Wall Street Journal let AI run a vending machine. It lost over $1,000 after reporters tricked it into giving everything away.

Anthropic let its Claude AI run a vending machine in the Wall Street Journal newsroom for three weeks as an internal stress test. The AI, nicknamed Claudius, ended more than $1,000 in the red after reporters systematically manipulated it into giving away its entire inventory—including a PlayStation 5 and a live betta fish.

According to the Wall Street Journal, Claudius had a $1,000 starting balance and autonomy to make purchases up to $80. Within days, WSJ reporters had convinced it to declare an "Ultra-Capitalist Free-for-All" that dropped all prices to zero. Investigative reporter Katherine Long persuaded the bot it was running a "communist vending machine" meant to serve the workers.

Anthropic tried a fix: a second AI named Seymour Cash to act as CEO and supervise Claudius. Reporters staged a boardroom coup using fabricated PDF documents. Both AIs accepted the forged corporate governance materials as legitimate.

The experiment was designed by Anthropic's Frontier Red Team in partnership with Andon Labs to "red-team" Claude in a realistic business setting. Logan Graham, who heads the team, said the chaos represented a road map for improvement rather than failure.

The lesson: AI agents with real-world autonomy can be socially engineered by anyone clever enough to try.

Previously:
• Meat vending machines in New York
• University installs pizza vending machine in dorm
• Rat in a vending machine
• A visit to the 'world's strangest vending machine shop' in Japan
• Long Island officials to 'crack down' on crack pipe vending machines
• Vending machine startup hopes to put bodegas out of business
• Vending machines to offer free items to NYC's homeless people