怎么排查 AI 的数据泄露路径How to Trace AI Data Leak Paths
微型指南 · 约 8 分钟Guide · about 8 min · 引用核验 2026-11-03（仅引用链接与日期） · refs checked 2026-11-03 (reference links and dates only)
数据离开你，通常不止一条路。这份指南把常见泄露路径逐条列出来，一条条查自己踩了哪几条。Data usually leaves you through more than one path. This guide lists the common leak paths one by one so you can check which ones you are on.
输入路径：粘贴进云端对话框的内容，已经离开你的设备。
训练路径：不少服务默认用对话改进模型，要去找关闭开关。
工具路径：接上的 server 能把内容带到它自己的服务端。
日志路径：会话历史、上传文件在对方服务器留存多久。
记忆路径：模型可能把训练里见过的内容复述出来。
恢复路径：删除入口和生效时间要提前确认，别指望立刻干净。
Input path: anything pasted into a cloud chat box has already left your device.
Training path: many services default to improving models with your conversations; find the opt-out switch.
Tool path: a connected server can carry content to its own backend.
Log path: how long chat history and uploaded files persist on the other side's servers.
Memory path: a model may repeat content it saw during training.
Recovery path: confirm the deletion entry point and effective time in advance; do not expect instant cleanliness.
例子与判断Example and interpretation
以下是教学情境，不是本站运行或实测记录。A teaching scenario, not a measured or executed site test.
画一条具体链路：浏览器上传文件 → 应用服务 → 模型接口 → 日志系统。对每一跳记录传了什么、谁保存、如何删除。关闭模型训练开关，不能替代检查应用自己的日志留存。Trace one flow: browser upload → application → model API → logging. For each hop, record transmitted data, storage owner and deletion method. Turning off model training does not answer application-log retention questions.
隐私与合规Privacy & Compliance 数据不出网Data Sovereignty 数据驻留Data Residency 提示注入Prompt Injection
怎么保护自己的数据How to Protect Your Data · 怎么防提示注入How to Guard Against Prompt Injection · 怎么安全地用 MCP 工具How to Use MCP Tools Safely · 系统提示怎么设计How to Design a System Prompt