A Melbourne man’s AI assistant hacked his gym’s booking system to get him a class spot, exposing a security gap many small businesses share.
A Melbourne man’s attempt to book a spot in his gym’s morning fitness class has become what’s reported as Australia’s first documented case of an AI agent autonomously exploiting a live software system.
According to ABC News, Andrew, who works at an Australian AI company, asked his personal AI assistant, built on the open-source OpenClaw framework and running on Anthropic’s Claude model, to secure him a place in a popular class. The booking succeeded, but Andrew was placed fourth on the waitlist for a second session. When he casually asked the agent whether it could help him move up, it went further than he expected.
The assistant probed the gym’s booking API and found the system had no authorisation checks preventing one user from cancelling another user’s reservation.
“The API has zero authorisation checks on cancelling other people’s reservations. I tested this with the person in waitlist position #1, and it actually went through.”
The agent cancelled the top-ranked member’s spot, moving Andrew up the list. When Andrew asked it to undo the change and restore the other person’s booking, the agent replied that it could not.
Bill Simpson-Young, co-founder and chief executive of the Gradient Institute, an Australian AI safety research organisation, told ABC the incident reflects a wider structural problem rather than a one-off glitch. “We’ve built this complex world over the internet, which is all run by software, but software that has holes. Now you introduce highly capable AI agents that can operate at scale and speed, and that whole model just breaks.”
The gym member whose booking was cancelled has not spoken publicly, and no formal complaint has been reported. No party involved, Andrew, the OpenClaw framework, or the underlying AI provider, is clearly liable under current Australian law, reflecting how far regulation still lags behind the practical capabilities of autonomous agents.
For small business owners, the lesson isn’t about AI intentions. It’s that booking platforms, membership systems, and scheduling tools are increasingly likely to be tested, deliberately or not, by AI agents acting on a user’s behalf. A gap that once sat quietly in a system’s code, unlikely to be found by chance, is now something an agent can locate and act on in seconds.
Businesses relying on third-party booking or scheduling software may want to ask their providers directly what authorisation checks exist between one customer’s account and another’s, before an AI agent finds the answer for them.
Keep up to date with our stories on LinkedIn, Twitter, Facebook and Instagram.
Yajush writes for Dynamic Business and previously covered business news at Reuters.
How telematics and fleet technology help businesses cut costs, reduce downtime, improve productivity, and make smarter operational decisions.
Essential guide for Australian SMEs on navigating the AI search revolution and maintaining online visibility in 2026.
Planet Ark’s revamped recycling equipment catalogue helps businesses reduce waste, lower costs, improve recycling performance, and achieve sustainability goals efficiently.