AWS Certified AI Practitioner AIF-C01 Practice Question

Guardrails for Amazon Bedrock provide two separate mechanisms for screening text. Content filters rely on built-in machine-learning classifiers that automatically detect broad categories such as hate, harassment, violence, or sexual content and block or modify the request or response when these categories are triggered. Because they work at the category level, developers do not have to enumerate individual words. Word filters, in contrast, require the developer to supply specific strings to look for, and PII redaction masks personally identifiable information rather than blocking sensitive topics. Model temperature is an inference-time generation setting and does not perform safety filtering. Therefore, enabling content filters is the correct approach for automatically blocking hateful or sexually explicit language.