AI and LLMs · Senior · Question 73 of 75
What is a practical first step to secure an existing LLM feature?
- 1Switch to the newest model and assume it's safer
- 2Threat-model its data, tools and output paths
- 3Rewrite the system prompt to be twice as long
Show the answer
Correct answer: Threat-model its data, tools and output paths
Explanation
Mapping data flows shows where injection, leakage or excessive agency can occur and which controls matter most.