Defense Layers Around the Model

Amazon Bedrock Guardrails is a configurable, model-agnostic safety layer you can attach to any inference call or run standalone via ApplyGuardrail (no model invocation needed). It ships six safeguard types:

• Content filters — hate, insults, sexual, violence, misconduct, and prompt-attack detection with per-category strength tuning.
• Denied topics — natural-language descriptions of subjects to block entirely.
• Word filters — exact-match custom word/phrase blocklists (e.g. competitor names, profanity).
• Sensitive information filters — probabilistic PII detection plus custom regex, with block or mask actions.
• Contextual grounding checks — flag RAG responses that aren't grounded in retrieved source or don't answer the query.
• Automated Reasoning checks — validate model claims against formal logical rules you define, surfacing hallucinations and unstated assumptions with mathematical certainty rather than a probabilistic guess.

Guardrails ship in Classic and Standard safeguard tiers (Standard extends detection into code elements like comments and identifiers), support cross-Region inference distribution, and can be enforced cross-account. Do not rely on model behavior alone for safety-critical decisions.