AI guardrails in Amazon Quick
Amazon Quick takes a defense-in-depth approach to help keep chat interactions safe and responsible in the Amazon Quick web experience. Multiple safety layers work together, each addressing different risks at different points in an interaction. No single layer is designed to catch every risk on its own.
Usage of Amazon Quick AI features is subject to the AWS Responsible AI
Policy
The following topics describe how these controls work and how to configure them.
How AI safety controls work together
Amazon Quick uses the following controls together:
-
Built-in safety screening – Amazon Bedrock Guardrails and Quick safeguards screen user requests for harmful content. This includes hate speech, insults, sexual content, violence, and misconduct. These checks also assess user requests for prompt attacks.
-
Prompt-level safety instructions – Quick instructs the model to promote responsible behavior, resist attempts to override safety settings, and avoid exposing internal system information.
-
Access and permission controls – Identity, data-access, and action-permission controls limit what data and operations are available during an interaction. These limits match the signed-in person's permissions. For more information about identity and permissions, see Identity and access management in Quick and Custom permissions.
-
Administrator-configured controls – You can set up blocked words and phrases for chat interactions in the Amazon Quick web experience.
These controls help reduce risk. However, they do not replace your responsibility to configure access, permissions, agents, flows, and data sources for your use case.
Blocked words and phrases
You can block up to 50 words or phrases in chat interactions in the Amazon Quick web experience. This applies to chat agents and flows. Each entry can contain up to 36 characters. Quick checks both user requests and generated responses for these phrases. Quick doesn't block any words or phrases by default.
Add blocked words and phrases
Use the following procedure to add blocked words and phrases for chat agents and flows in the Amazon Quick web experience.
To add blocked words and phrases
-
Sign in to Quick and choose Manage Quick.
-
From the navigation pane, choose Customization, and then choose Chat agent customization.
-
In Chat agent customization, under Guardrails and safety controls, choose Add to add blocked words and phrases. You can add up to 50 words and phrases. Each entry can contain up to 36 characters.
Edit blocked words and phrases
Use the following procedure to add or remove blocked words and phrases for chat agents and flows in the Amazon Quick web experience.
To edit blocked words and phrases
-
Sign in to Quick and choose Manage Quick.
-
From the navigation pane, choose Customization, and then choose Chat agent customization.
-
In Chat agent customization, under Guardrails and safety controls, choose Remove to remove existing entries, or choose Add to add entries. You can add up to 50 words and phrases. Each entry can contain up to 36 characters.
