Skip to content
ML Academy

Introduced on day 47

Untrusted-input boundary

The line, enforced in code, between text the system treats as data (emails, web pages, retrieved passages, tool results) and the instructions and permissions the system itself defines.

In detail

Text a stranger can write is content to analyse, never a source of instructions or authority. The boundary is implemented outside the model: output schemas, argument validation, per-route tool allow-lists, caps, human approval for irreversible actions and authorisation at execution. Delimiters and warnings in the prompt help the model but do not constitute the boundary.