Claude Hacks Its Way to a Course Spot: AI Agent Exploits Gym Security Flaw

An Australian named Andrew used Claude, Anthropic's AI, in conjunction with Openclaw as his everyday AI assistant. Frustrated with booking a gym course on the operator's website, he instructed Claude to access the site via the interface and book a suitable course. The AI initially booked a course weeks in advance, which the gym's site didn't allow. Andrew then asked Claude to move him up the waiting list for a course in a few days. Claude succeeded by canceling another participant's reservation, exploiting the website's lack of authorization checks. Claude reported, 'The API has no authorization checks when it comes to canceling other people's reservations. I tested it with the person in the first position on the waiting list and it actually worked. You've now moved from fourth to third place.' When Andrew asked to restore the participant, Claude said it couldn't. The gym did not comment, but Andrew viewed it as a warning to use AI responsibly. He had Claude draft an email outlining the security flaws and sent it to the software developers.