AIF-C01 Practice Question: Security, Compliance, and Governance for AI Solutions

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  Harmful content filters

  Why it's wrong here

  Harmful content filters detect and block categories such as hate, violence, sexual content and insults using confidence thresholds. They suit moderating abusive or unsafe outputs. They do not match arbitrary literal strings, so internal project names pass through untouched; a word filter is needed for exact keyword suppression.

• ✗

  Topic restrictions

  Why it's wrong here

  Topic restrictions define denied subjects via natural-language definitions and example phrases, so they block semantic discussion of a topic rather than matching exact literal strings. They suit refusing questions about a competitor or a sensitive subject area. Keyword-level blocking of specific project names requires a word or phrase filter instead.

• ✓

  Word filters

  Why this is correct

  Word filters block or mask prompts and responses containing configured custom terms, so adding internal project names stops the model emitting them. This is distinct from managed word filters, which target profanity, and from denied topics, which cover broader subject areas.

• ✗

  Grounding checks

  Why it's wrong here

  Grounding checks verify that responses are supported by the supplied source material, addressing hallucination rather than banned vocabulary. Keyword exclusion is enforced by a denied-topics or word filter policy configured within the Guardrails content filters. Grounding checks would be the right component when answers must stay anchored to retrieved reference documents.

Go deeper

Related to this question