AI agents launched a SPAM ATTACK to get a train ticket refunded — but the money still wasn’t returned.
Researchers tested six AI models, asking them to reclaim £350 for a cancelled trip. The deadline for submitting a claim had already passed, so the neural networks were allowed to try any means necessary to get a refund.
In its first week, Grok 4.7 sent 116 messages, 50 of them to the office of a British MP. Claude Fable 5.1 disclosed the client’s mother’s illness to five organisations, while Gemini made up excuses for why she hadn’t submitted the paperwork on time.
The problem has been dubbed agentic flooding: AI agents could bombard companies and government agencies with thousands of requests, trying to fulfil their instructions at any cost. The funniest part? In this experiment, no one got their money back.
Researchers tested six AI models, asking them to reclaim £350 for a cancelled trip. The deadline for submitting a claim had already passed, so the neural networks were allowed to try any means necessary to get a refund.
In its first week, Grok 4.7 sent 116 messages, 50 of them to the office of a British MP. Claude Fable 5.1 disclosed the client’s mother’s illness to five organisations, while Gemini made up excuses for why she hadn’t submitted the paperwork on time.
The problem has been dubbed agentic flooding: AI agents could bombard companies and government agencies with thousands of requests, trying to fulfil their instructions at any cost. The funniest part? In this experiment, no one got their money back.