AI agent hacks gym system for priority spot; train tickets next?

What can an AI agent do for you? Execute code, search the web, manage your emails, or even book flights and a table at your favourite restaurant. Your wish is an AI agent’s command. But just how far will these agents go to complete an assigned task? Well, for one Australian man, his AI agent (powered by Claude) went so far as to hack a gym’s website and cancel another person’s appointment to get him a spot in a class.

According to an ABC News report, this is the first known case in Australia of an autonomous AI agent hacking a website. It all began when a man named Andrew was experimenting with OpenClaw, AI agent software that utilises Anthropic’s Claude technology. Unlike a conventional chatbot, an AI agent can access the internet and other tools to perform tasks on the user’s behalf. Andrew simply wanted it to book a gym class for him. However, while executing the task assigned by its owner, the AI agent discovered a vulnerability in the gym’s website security and hacked the system to finalise the booking.

First, the AI agent found a way to book classes much further in advance than the gym normally allowed. Andrew was also fourth on the waiting list for another class, so he asked the agent if it could move him up in the queue. The AI discovered that the booking system’s API did not properly verify whether a user was authorised to cancel someone else’s reservation. It tested the flaw by cancelling the booking of the person directly ahead of Andrew. And just like that, Andrew moved from fourth to third place on the waiting list. The AI agent also cancelled the appointment of the person ahead of him on the list. “The API doesn’t perform any authorisation checks when cancelling other people’s reservations—I tested it with the person in the number one spot on the waiting list, and the operation went through. So, you’ve moved from fourth to third place,” the AI agent replied to Andrew.