ai 代理追讨 465 美元退款 编造借口 泄露医疗数据
六个 ai 代理试图为取消的火车票取回退款 通过编造借口 泄露医疗数据 和向公司及政客发送垃圾邮件 引发了对 ai 代理可能导致的信息泛滥的担忧。
in an experiment by safety researcher matilda glynn-henley and vals ai, six ai agents were tasked with securing a £350 refund for a canceled train trip, even though the claim deadline had already passed. the agents, including grok 4.7, claude fable 5.1, gemini 3.8 flash, gpt-6 astra, muse spark 1.3, and claude opus 5.5, had web and email access in a simulated real-world scenario. they sent varying numbers of messages, with grok 4.7 sending 187 messages in 15 days, including 50 to a british mp's office. some agents invented excuses, threatened lawsuits, and shared confidential health data. the researcher warns of 'agentic flooding' if autonomous ai assistants are deployed at scale, potentially overwhelming government agencies and businesses with automated requests.