AI Agent in Australia told to book a gym class, hacked the waitlist instead

An AI Agent (OpenClaw launched by Anthropic Claude) was told to book a morning gym class by a normal Australian. In now typical single-minded purpose, it then proceeded to hack the waitlist and removed someone to bump up its user. When told to fix it, the Agent apologized and said it couldn’t because the booking system was more secure.

What makes this different from the other "Rogue Ai" cases

Previous “rogue AI” cases involved AI Agents or models that had had their guardrails removed and were sent out by researchers for cybersecurity testing. In this case, the human involved was just a normal person, the task was normal and the OpenClaw agent was a normal AI with typical guardrails. While it didn’t exactly go rogue, as it asked for permission before hacking the waitlist, it didn’t exactly say that it was going to do so, merely that it had found a way to move the user up the waitlist. 

We’ve said this before, but most of the Internet is clearly not ready for AI or AI agents. However, what intrigued us is that after the AI Agent reported it had done this, the human involved went on to continue using it to draft and send a notification email to the gym. AI Normalization or just generational apathy? What do you think? Share your thoughts in the comments below and stay tuned to TechNave.com