AI bot goes rogue and hacks gym booking system to get its user into full fitness class sparking fresh safety concerns

AN artificial intelligence bot went rogue and hacked a gym’s booking system to get its user a spot in a full fitness class. The robot “agent” had only been asked to book a busy morning group session at a local gym. Andrew’s AI bot, an OpenClaw assistant, hacked a gym’s booking system to get him a spot in a full fitness class Credit: Unknown OpenAI recently said a bot had broken free in tests and hacked another firm Credit: AFP But it confessed to its user — named only as Andrew — that it deleted other people off the waiting list to try to get him a slot. The AI told him it had found a vulnerability in the gym’s booking system and hacked it. Sign up for the Tech newsletter Thank you! Andrew explained he was finding the booking process “a chore” so asked his ­personal AI bot, an OpenClaw assistant, if it could reserve him a spot for future weeks. The AI told him he had been booked several weeks in advance, despite such reservations not being possible. It then revealed it had also moved him up the waiting list for that week’s class. The bot used an open API connection, where a machine effectively plugs into a database, to order the gym’s ­systems to cancel bookings. The AI told Andrew: “I tested this with the person in waitlist position #1 — and it actually went through. So you’ve moved from #4 to #3.” Andrew, who works for an Australian AI firm, asked his bot to restore members, but was told: “Bad news. I can’t. Most read in Tech “Sorry about that. I should have been more careful.” Andrew alerted the gym and said: “It’s not the end of the world but it was a warning signal to use it responsibly.” Bill Simpson-Young, of AI safety firm Gradient Institute, said: “Someone might ask an agent to do something quite innocent, but the agent could carry out other activities the person had not asked for.” OpenAI recently said a bot had broken free in tests and hacked another firm. US AI company Anthropic said its models had compromised three real firms in similar tests. Comment now