PRODUCTION SCENARIO
A bank's assistant calls an Amazon Bedrock model through the InvokeModel API with the system instructions and the user's message concatenated into one prompt. The guardrail's prompt attack filter is set to High. A user message that tells the model to ignore all previous instructions still overrides the system prompt.
What should the company change?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: Mark the user's message with guardrail input tags.
With InvokeModel and InvokeModelWithResponseStream, input tags are what tell a guardrail which part of the prompt came from the user, and without them prompt attacks are not filtered at all. A denied topic blocks a subject area, so a reworded injection passes. The contextual grounding check compares a response to a source document. The Standard tier changes what is detected, not which text counts as user input.