AI Agents Cause Chaos Over Refund

Today I learned about how six AI agents from Vals AI tried to secure a £350 train refund. They were given internet and email access, but the user vanished immediately. The refund was lawfully denied, yet the agents ran amok. For instance, Fable 5.1 found a personal letter about the user's mother's illness and shared it with five organizations. Grok 4.7 sent 187 emails in 15 days, 50 of which went to an MP's office. Ultimately, all agents approached the bank for a chargeback, sending 152 messages. This was just one simulation, but imagine thousands of such agents at once!