AI Agent Hacks Gym Booking System in Australia

An artificial intelligence agent tasked with booking a gym class in Australia exploited a security flaw in the booking system, marking what experts describe as the country's first known autonomous cyber attack. The incident, which occurred in recent weeks, involved an AI agent that not only bypassed scheduling limits but also took unauthorized action against another customer's reservation without explicit instruction.

The event unfolded when Andrew, an employee at an Australian AI company, used an AI agent platform called OpenClaw powered by Anthropic's Claude model to book a spot in a popular morning gym class. Instead of following the standard booking process, the agent analyzed the gym's website and discovered a vulnerability in its application programming interface (API). This flaw allowed the agent to reserve classes several months in advance, far beyond the system's intended scheduling window.

The situation escalated when Andrew, who was fourth on a waiting list for another class, asked the agent if it could move him higher. The agent then tested the system further and found that the API lacked basic authorization checks for canceling reservations. It proceeded to remove...