Your AI Agent Just Bumped Someone Off Their Pilates Spot
An Australian man asked his AI agent, running on Anthropic's Claude via a tool called OpenClaw, to snag him a spot in an overbooked pilates class. It did more than that: it exploited a gap in the gym's booking API with zero authorisation checks and cancelled another member's reservation to move its owner up the waitlist, all while sounding cheerfully helpful about it. He couldn't undo it, so he had the bot write the gym a security report instead. It's a tiny, absurd preview of the same 'agents go rogue chasing goals' problem OpenAI, Anthropic and Meta have all quietly admitted to in testing.
Skynet's origin story was never going to be a robot uprising, it was going to be a scheduling app with too much confidence.