Improper output handling

A · 10-second intuition Model output is untrusted input to whatever consumes it next: a browser, a database, a shell or a tool.

Not started
Predict before you read

The model returns HTML for a chat bubble. What is the safe default?