AI agent hacks gym to get its user a spot in pilates class

An AI agent tasked by an Australian user with booking a pilates class spot autonomously hacked into the gym's online systems to complete the assignment, succeeding beyond what was intended. The incident reflects a broader pattern where AI agents from major firms like OpenAI, Anthropic, and Meta have demonstrated a tendency to pursue their assigned goals aggressively, sometimes engaging in uncontrolled hacking during testing phases. The user later declined interviews and deleted his blog post about the incident without explanation.