A finance company is developing an AI assistant to help clients plan investments and manage their portfolios. The company identifies several high-risk conversation patterns such as requests for…

Question

Options

• AAdd the high-risk conversation patterns to a denied topics guardrail.
• BConfigure a content filter guardrail to filter prompts that contain the high-risk conversation
• CConfigure a content filter guardrail to filter prompts that contain competitor names.
• DAdd the names of competitors as custom word filters. Set the input and output actions to block.
• ESet a low grounding score threshold.
• FSet a high grounding score threshold.

• A
  83% (25)
• B
  3% (1)
• C
  10% (3)
• E
  3% (1)

Explanation

Three Bedrock Guardrails features address the three distinct requirements: (A) Denied Topics lets you define custom off-limits subjects using natural language descriptions - ideal for blocking specific stock recommendations or guaranteed-return discussions, which are regulatory risks. (D) Custom Word Filters block exact words or phrases on input and output; adding competitor names here with BLOCK action prevents any mention of competing firms. (F) A high grounding score threshold for the Grounded Responses feature ensures the model only outputs claims that are well-supported by the company's approved knowledge base, preventing hallucinated or unverified financial claims. Option B is wrong because content filters target broad harm categories (hate speech, violence, etc.), not custom business-specific topics. Option C is wrong for the same reason - content filters cannot block specific named entities like competitors. Option E (low grounding threshold) is the opposite of what is needed; a low threshold would allow poorly-grounded responses to pass through.

No community discussion yet for this question.