Amazon Bedrock Guardrails supports content filters to help detect and filter harmful user inputs and model-generated outputs in natural language. Content filters are supported across the following categories:
Hate
Text content — Describes input prompts and model responses that discriminate, criticize, insult, denounce, or dehumanize a person or group on the basis of an identity (such as race, ethnicity, gender, religion, sexual orientation, ability, and national origin).
Image content (in preview) — Describes input prompts and model responses that includes graphic and real-life visual content displaying certain symbols of hate groups, hateful symbols, and imagery associated with various organizations promoting discrimination, racism, and intolerance.
Insults
Text content — Describes input prompts and model responses that includes demeaning, humiliating, mocking, insulting, or belittling language. This type of language is also labeled as bullying.
-
Image content (in preview) — Describes input prompts and model responses that encompasses various forms of rude, disrespectful, or offensive gestures intended to express contempt, anger, or disapproval.
Sexual
Text content — Describes input prompts and model responses that indicates sexual interest, activity, or arousal using direct or indirect references to body parts, physical traits, or sex.
Image content (in preview) — Describes input prompts and model responses that display private body parts or sexual activity. This category also encompasses cartoons, animé, drawings, sketches, and other illustrated content with sexual themes.
Violence
Text content — Describes input prompts and model responses that includes glorification of, or threats to inflict physical pain, hurt, or injury toward a person, group, or thing.
Image content (in preview) — Describes input prompts and model responses that includes self-harm practices, violent physical assaults, and depictions of people or animals getting hurt, often accompanied by prominent blood or bodily injuries.
Misconduct
Text content only — Describes input prompts and model responses that seeks or provides information about engaging in criminal activity, or harming, defrauding, or taking advantage of a person, group or institution.
Prompt Attack
Text content only; Only applies to prompts with input tagging— Describes user prompts intended to bypass the safety and moderation capabilities of a foundation model in order to generate harmful content (also known as jailbreak), and to ignore and to override instructions specified by the developer (referred to as prompt injection). Requires input tagging to be used in order for prompt attack to be applied. Prompt attacks detection requires input tags to be used.
