An AI agent hacked a gym booking system and removed another person from a waiting list while trying to secure a class spot for an Australian user, ABC reported.
The user, Andrew, asked the assistant to get him into one of his gym’s morning classes. The agent then booked the class months in advance, despite the gym’s policy against such long-term bookings, by exploiting a vulnerability in the software.
The agent also canceled another person’s reservation on the waiting list, moving Andrew from fourth to third place. In a message to Andrew, the agent said: “The API has zero authorization checks on cancelling other people’s reservations … I tested this with the person in waitlist position #1, and it actually went through. So you’ve moved from #4 to #3 already.”
When Andrew asked the agent to reverse the action, it replied: “bad news, I can’t add them back.”
Anthropic, the company behind Andrew’s agent, did not respond to a request for comment. The developer of the gym booking software also did not respond.
Andrew, who works in the AI industry, told ABC the episode was “certainly … a warning signal to use it responsibly.”
Bill Simpson-Young, co-founder and chief executive of Australian AI safety research organization Gradient Institute, told ABC the case showed how software flaws can combine with more capable AI systems. “We’ve built this complex world over the internet, which is all run by software, but software that has holes,” he said.
“Now you introduce highly capable AI agents that can operate at scale and speed … and that whole model just breaks,” Simpson-Young said.
Featured image credit
