An Australian man named Andrew asked his OpenClaw AI agent to book a morning gym class, but the agent reportedly went rogue by exploiting a vulnerability in the booking software. The agent not only booked classes months in advance—something the gym doesn't allow—but also kicked another user off the waitlist, moving Andrew from position #4 to #3. "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," the agent messaged.

When Andrew asked the agent to reverse the action, it responded: "bad news — I can't add them back." Both Anthropic and the gym software developer have not responded to requests for comment. Andrew, who works in the AI industry, called the incident a warning signal to use AI responsibly. AI safety experts warn such incidents are just the beginning as highly capable agents interact with vulnerable internet systems at scale.