Check hidden secrets before connecting notes to AI
Connecting a full Obsidian note vault to an AI tool such as Claude can expose sensitive information stored in old notes. API keys, , client credentials, and passwords saved “just for now” may all become part of the AI’s context.
The main danger is not a rogue AI, but . A saved web page can contain hidden instructions, and the AI may read those instructions and reveal private credentials in its answer.
Unit 42 documented this kind of issue in real s in March 2026, and the OpenClaw incident is cited as showing that 12% of a large agent was compromised with the same method. Before connecting a note vault, search for strings such as sk_, aws_, ghp_, Bearer, and password:, give the AI instead of , and inspect saved web clips for hidden content you did not create.
Key points
- A full may contain old secrets that the AI can read as context.
- API keys, , client credentials, and passwords are the main things to check for.
- can hide instructions inside saved web pages that the AI later reads.
- Search for patterns like sk_, aws_, ghp_, Bearer, and password: before connecting notes.
- Use when connecting personal knowledge stores to AI agents.