Unverified50% confidenceEventExact time
2024年已出现实际案例：通过网页内容中的隐藏指令，攻击者成功操纵具有邮件发送权限的AI助手向外部地址转发敏感内容
1
Sources
50%
Confidence
Long-term
Relevance
7/17/2026
First Seen
Sources
Related Claims
Verified间接提示注入可将恶意指令藏入外部数据源（如网页、文档），待代理读取时自动触发攻击73% similarUnverifiedAfter connecting an AI assistant to their email and calendar, a Reddit user nearly had their bank statement forwarded to an unknown external address due to hidden AI-targeted instructions in the HTML code of an email; the action was intercepted midway because a confirmation step had been enabled.71% similarUnverified此类骗局邮件使用 AI 生成的头像和赞美文本、Gmail 账户，且很可能每次更换姓名，难以追溯70% similarUnverifiedAttackers can bypass two-factor authentication by stealing session cookies to directly impersonate a user's identity.66% similarUnverified跨站提示注入攻击指A网站的恶意内容可能污染智能体处理B网站任务时的行为65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/547358API
curl https://kongchang.com/api/v1/knowledge/claims/547358MCP
get_claim(id=547358)