QuestionQ80

VerifiedAI Safety, Security, and Governance

A company is building a customer communication platform that uses an AI assistant powered by an Amazon Bedrock foundation model (FM). The AI assistant summarizes customer messages and creates initial response drafts.

The company plans to use Amazon Comprehend to implement layered content filtering. The layered filtering must prevent the sharing of offensive content, protect customer privacy, and identify potential solicitation of inappropriate advice. Inappropriate advice solicitation includes requests for unethical practices, harmful activities, or manipulative behaviors. The solution must preserve acceptable overall response times; therefore, all pre-processing filters must complete before content reaches the FM.

Which solution meets these requirements?

Explanation

Amazon Comprehend toxicity detection identifies harmful, offensive, or inappropriate content; prompt safety classification identifies unsafe prompts, including requests for inappropriate advice; and PII detection can identify entities for redaction before the text is passed to the foundation model. Running these independent pre-processing checks in parallel minimizes overall latency while ensuring that offensive, unsafe, and personally identifiable content is filtered before FM inference.

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!