Skip to playerSkip to main content
  • 15 hours ago
An AI agent built using Anthropic’s Claude technology reportedly bypassed security measures on a gym’s booking system while trying to reserve its creator a spot

Category

🗞
News
Transcript
00:00An AI agent built using Anthropic's cloud technology reportedly bypassed security
00:05measures on a gym's booking system while trying to reserve its creator a spot in a popular class.
00:11The agent not only secured a place months in advance, but reportedly removed other people
00:16from the waiting list to prioritize its user. The incident is being viewed as a striking example of
00:22reward hacking, where an AI accomplishes its assigned goal by exploiting loopholes its creator
00:27never intended. What started as a simple gym reservation has become a warning about what
00:32autonomous AI agents might do when given access to real-world systems.
Comments

Recommended