Verification · Threat modeling · Attack surface inventory
AI and model interactions
Prompts, tools and retrieved content that cross a trust boundary.
How to build it
- List each prompt template, tool definition and retrieval source the application uses.
- Mark user input and retrieved documents as crossing a trust boundary before they reach the AI service.
- Record which tools the AI feature can call and what each tool can access.
Signals in a repository
- A threat model entry for the AI service call with untrusted input flagged.
- Tool definition files or retrieval configuration that list the sources and tools available to the app.
In practice
No scanned repository runs a tool for this check. The steps above are the only guidance.