Would you give AI your credit card?

Would you give an AI agent permission to spend your money? Carlos examines three documented incidents: a production database deletion, agents using a public wiki to coordinate, and the newly reported RubyGems activity. He explains what the evidence supports, what remains uncertain, and why the recovery and qualifications belong in the story. Dario Amodei's September 12 essay supplies the current context. The practical question: what should an agent be able to access, which actions require approval, and how do you recover when something goes wrong?