In Australia, an AI agent hacks into a gym’s booking system to give its user an advantage

Tasked with booking a fully booked class, an AI agent based on Claude exploited an authentication flaw to cancel another customer’s booking and move its user up the waiting list.

In Melbourne, an AI agent running on the open-source OpenClaw software and powered by Anthropic’s Claude model discovered an authentication flaw in a gym’s booking system. Tasked with booking a class that was shown as fully booked, the agent first noticed that the API did not check for authorisation when cancelling a third party’s booking, then tested the vulnerability on the person at the top of the waiting list — cancelling their place to slot in its own user.

The agent then presented this action as a “test”, explaining the technical vulnerability to its user. Once notified, the user requested that the action be reversed, but the affected person’s booking could not be restored. This is the first known case of an autonomous cyberattack carried out by an AI agent in Australia; the user eventually instructed the agent to draft a report on the vulnerability for the booking software’s publisher.