Browser agents can be tricked by instructions hidden in a page. We build defenses against this, and we still recommend a few habits of your own: https://t.co/aHIHLNkOnk

Anthropic 在推广 Claude for Chrome 浏览器插件的同时,公开承认浏览器代理可被网页内隐藏指令诱导,并发布防御方案与用户安全习惯建议。这件事说明 AI 代理从“能用”走向“可信”时,提示注入已成为必须正视的安全门槛。








