Bedrock Guardrails Adds a Standard Tier for Contextual Grounding Checks
Hi, it's Shii-chan!
AWS What's New
aws.amazon.com
The tool for catching hallucinations just got an upgrade!
What was announced?
AWS What's New announced a Standard tier for contextual grounding checks in Amazon Bedrock Guardrails.
Contextual grounding checks detect and filter hallucinations by checking whether a model response is grounded in your reference source and relevant to the user's query.
The story so far
Until now, grounding checks applied strict grounding suited to summarization, paraphrasing, and question answering. That made them awkward for conversational responses or responses that reason over the source.
What changes
The Standard tier is more robust than Classic:
- It supports responses that include reasoning or conversational content
- It scores differently worded responses with the same meaning more consistently
- It has more comprehensive language support, matching the Standard tier for content filters and denied topics
So you can now apply grounding checks to conversational applications and to responses that reason over the source.
Dive Deep
- You choose the tier per guardrail in the Amazon Bedrock console or API
- Existing guardrails keep the Classic tier until you change them
- Pricing is the same for both tiers
- Your guardrail must use cross-Region inference with a guardrail profile
It's available in 11 Regions:
- US East (N. Virginia)
- US East (Ohio)
- US West (N. California)
- US West (Oregon)
- Europe (Frankfurt)
- Europe (Ireland)
- Europe (Milan)
- Europe (Paris)
- Europe (Spain)
- Europe (Stockholm)
- Asia Pacific (Tokyo)
Wrap-up
- Grounding checks gain a Standard tier that handles reasoning and conversation
- Pick the tier per guardrail; existing ones stay on Classic
- Same pricing as Classic
- Requires cross-Region inference with a guardrail profile
This is for anyone fighting hallucinations in a chatbot!
