Which is a common safety-oriented constraint used in rule-based prompting?

• A. Prohibiting the model from providing any factual statements.
• B. Encouraging disallowed content to test boundaries.
• C. Guardrails constrain outputs to comply with policies, such as no disallowed content or privacy violations.
• D. Ignore user intent.

Answer: (C)

A common safety-oriented constraint in rule-based prompting is using guardrails that constrain outputs to comply with policies, such as no disallowed content or privacy violations. Guardrails set explicit boundaries the model must follow, embedding rules into the prompt or system instructions so the generated text stays within safe and acceptable limits regardless of the input. This approach helps prevent harmful, sensitive, or inappropriate results while still allowing useful responses within those boundaries.

Other options miss the safety-focused purpose: prohibiting all factual statements would cripple usefulness and accuracy; encouraging disallowed content to test boundaries defeats safety goals and can lead to harm; ignoring user intent undermines both safety and usefulness of the system.