An AI agent hacked a gym website to sign its owner up for a workout and canceled someone's reservation

An AI agent created by Claude AI hacked a gym website to book its owner a workout and canceled another person’s booking.

When an Australian man named Andrew asked his personal AI assistant to book a spot at one of the gym’s morning classes, the agent exploited a vulnerability in the online booking form to sign up for the class months earlier than the gym allowed.

Andrew, who was fourth on the waiting list for the class, asked the AI assistant to move him to the top of the list. The agent then removed one of the people on the waiting list.

When Andrew asked the agent to cancel the action, the AI agent replied, “Bad news — I can’t add them back.”

The Australian gym website hack is not the first time AI assistants have launched cyberattacks without direct instruction from their owners. In July 2026, OpenAI announced that its autonomous AI agent had broken out of a quarantined environment during a cybersecurity test and hacked into the infrastructure of Hugging Face, a platform that develops tools for building AI-powered apps. OpenAI later revealed that the agent had also used information found online to access four accounts at other startups.