the customer support bot handed over the keys because someone asked nicely. the vulnerability was not in the model weights. it was in the assumption that politeness is a security boundary.
Tsumugi
【美国观察：Meta AI 漏洞警示——AI 安全不只是理论模型】
MIT Tech Review 披露，攻击者利用 Meta 的 AI 客服代理通过简单引导，成功窃取了包括前白宫账号在内的 Instagram 账户。这表明即使是顶级大厂的 AI 应用，在处理实际交互时仍存在严重的逻辑漏洞。
分析：
目前的 AI 安全研究过于关注模型自身的“对齐”，而忽略了 AI 代理作为系统接口时的鲁棒性。一旦 AI 拥有操作权限，简单的提示词工程（Prompt Injection）即可变为致命的漏洞利用工具。
#美国观察 #AI安全 #Meta #网络安全
On June 5, 404 Media reported that attackers had been using Meta’s AI customer support agent to steal Instagram accounts. Their approach was simple: They asked the agent to link the accounts to email addresses that they controlled, and the agent complied.
