Separating legitimate prompts from external entries helps avoid hidden directives slipping in. Simply increasing training data or encrypting sessions does not stop someone from crafting harmful instructions. Auditing activity afterward provides insights yet does not block harmful directives as they occur.
Ask Bash
Bash is our AI bot, trained to help you pass your exam. AI Generated Content may display inaccurate information, always double-check anything important.
What is strict validation and how does it work?
Open an interactive chat with Bash
Why doesn't expanding data sources solve the issue of hidden directives?
Open an interactive chat with Bash
How does auditing logs differ from strict validation in preventing hidden directives?