AWS Guardrails

AWS Guardrails provides configurable security measures to help safely build generative AI applications at scale. It uses automated reasoning to help prevent factual errors caused by hallucinations. It

AWS Guardrails features include:

• Content filtering (preventing explicit, violent content, etc.)

• Topic deviation prevention

• Word restrictions

• Sensitive data removal

• Hallucination prevention

Operation Steps

Step 1: Configure Settings (Name, Description, Blocked Message Response)

Example settings:

Step 2: Set Content Filtering (User Input, Assistant Response)

You can specify harm levels, including image inputs

Example settings:

Step 3: Topic Rejection, Preventing Off-Topic Responses

You can prevent off-topic responses by adding descriptions, definitions, and examples

Example settings:

Step 4: Word Restrictions (Blocking Profanity or Specific Terms)

For example: To prevent the model from mentioning Voldemort, add words - multiple words can be input at once (csv, txt, ctrl + c).

Example settings:

Step 5: Sensitive Data (PII) Blocking and Regex Replacement

Several patterns are preset, such as credit cards, vehicle numbers, ID numbers, etc.

Example settings:

Regex replacement allows you to replace unwanted structures in AI assistant responses using desired regular expressions.

Step 6: Preventing Hallucinations and Setting Relevance Threshold for Response Blocking

Example settings: