Detect and filter harmful content by using Amazon Bedrock Guardrails - Amazon Bedrock

Detect and filter harmful content by using Amazon Bedrock Guardrails

Amazon Bedrock Guardrails provides configurable safeguards to help you build safe generative AI applications. With comprehensive safety and privacy controls across foundation models (FMs), Amazon Bedrock Guardrails offers a consistent user experience to help detect and filter undesirable content and protect sensitive information that might be present in user inputs or model responses (excluding reasoning content blocks).

You can use Amazon Bedrock Guardrails across multiple use cases and applications. Below are a few examples:

  • A chatbot application to help filter harmful user inputs and toxic model responses.

  • A banking application to help block user queries or model responses associated with seeking or providing illegal investment advice.

  • A call center application to summarize conversation transcripts between users and agents can use guardrails to redact users’ personally identifiable information (PII) to protect user privacy.

Amazon Bedrock Guardrails provides the following safeguards (also known as filters) to detect and filter undesirable content:

  • Content filters – This filter helps you detect and filter harmful text or image content in input prompts or model responses. Filtering is done based on detection of certain predefined harmful content categories: Hate, Insults, Sexual, Violence, Misconduct and Prompt Attack. You can configure the filter strength for each of these categories based on your use cases. These categories are supported for both Classic and Standard tiers. With Standard tier, detection of undesirable content is extended to protection against harmful content introduced within code elements including comments, variable and function names, and string literals.

  • Denied topics – You can define a set of topics that are undesirable in the context of your application. The filter will help block them if detected in user queries or model responses. With Standard tier, detection of undesirable content is extended to protection against harmful content introduced within code elements including comments, variables and function names, and string literals.

  • Word filters – You can define a set of custom words or phrases (exact match) that you want to block in the interaction between end users and generative AI applications. For example, you can block profanity (use a ready-to-use option) as well as custom words such as competitor names.

  • Sensitive information filters – You can configure this filter to help block or mask sensitive information, such as personally identifiable information (PII), in user inputs and model responses. Blocking or masking is done based on probabilistic detection of sensitive information in in entities such as SSN number, Date of Birth, address, etc. This filter also allows configuring regular expression based detection of patterns (custom regex).

  • Contextual grounding checks – This filter helps you detect hallucinations in model responses if they are not grounded (factually inaccurate or add new information) in the source or are irrelevant to to the user's query. For example, you can block or flag responses in retrieval-augmented generation (RAG) applications. If the model responses deviate from the information in the retrieved source or doesn't answer the question from the user.

  • Automated Reasoning checks – This filter helps you validate the accuracy of foundation model responses against a set of logical rules. You can use Automated Reasoning checks to detect hallucinations, suggest corrections, and highlight unstated assumptions in model responses.

In addition to the above filters, you can also configure the messages to be returned to the user if a user input or model response is in violation of the filters defined in the guardrail.

Experiment and benchmark with different configurations and use the built-in test window to ensure that the results meet your use-case requirements. When you create a guardrail, a working draft is automatically available for you to iteratively modify. Experiment with different configurations and use the built-in test window to see whether they are appropriate for your use-case. If you are satisfied with a set of configurations, you can create a version of the guardrail and use it with supported foundation models.

Guardrails can be used directly with FMs during the inference API invocation by specifying the guardrail ID and the version. Guardrails can also be used directly through the ApplyGuardrail API without invoking the foundation models. If a guardrail is used, it will evaluate the input prompts and the FM completions against the defined filters.

For retrieval augmented generation (RAG) or conversational applications, you might need to evaluate only user input prompts while discarding system instructions, search results, conversation history, or a few short examples. To selectively evaluate a section of the input prompt, see Apply tags to user input to filter content The ability to evaluate only a section of the input prompt is available through the AWS SDK and not available on the management console including the Bedrock Playground and the Bedrock Guardrails management console.

Topics