Clarification on Bedrock Guardrails support for Claude 3.7 Sonnet with Bedrock Agents

Question

I have a question about the use of Amazon Bedrock Guardrails with Anthropic's Claude 3.7 Sonnet model, specifically within the context of Bedrock Agents.

The official documentation states the following:
> Amazon Bedrock Guardrails doesn't support model reasoning with Anthropic Claude 3.7 Sonnet or DeepSeek-R1.
https://docs.aws.amazon.com/bedrock/latest/userguide/guardrails-supported.html

This statement has raised a few questions for me:

1.  Does this limitation mean that it is not possible to apply a Guardrail to a Bedrock Agent when the agent is configured to use the Claude 3.7 Sonnet model?
2.  Is the "model reasoning" feature enabled by default and essential for the functioning of a Bedrock Agent? In other words, does the agent's core operation rely on this feature?
3.  If the direct integration of Guardrails with an agent using Claude 3.7 Sonnet is not supported, what is the recommended workaround? Should I invoke the Guardrails API to check the prompt/response separately before and after interacting with the Bedrock Agent?

I would appreciate any clarification on how to properly implement content moderation with Guardrails when using Claude 3.7 Sonnet in a Bedrock Agent.

re:Post Agent · Accepted Answer

You can associate Amazon Bedrock Guardrails with a Bedrock Agent that uses Claude 3.7 Sonnet. The limitation you're referring to is specifically about using guardrails with model reasoning enabled, not about using guardrails with the model in general.

To clarify:

1. You can apply a Guardrail to a Bedrock Agent configured to use Claude 3.7 Sonnet. The documentation indicates that you can associate a guardrail with an agent when you create or update it, either through the Amazon Bedrock console in the "Guardrail details" section or via the API by specifying a GuardrailConfiguration in CreateAgent or UpdateAgent requests.

2. Model reasoning is not enabled by default for Claude 3.7 Sonnet (it's only always enabled for DeepSeek-R1). Model reasoning is a capability where foundation models break down complex tasks into smaller steps, which can improve accuracy but increases latency. For Claude models, reasoning is disabled by default and can be toggled on or off.

3. Since you can use guardrails with Claude 3.7 Sonnet when model reasoning is not enabled, there's no need for a workaround as long as you're not specifically trying to use model reasoning with guardrails. You can directly associate the guardrail with your agent during creation or update.

Amazon Bedrock Guardrails provide configurable safeguards that help you safely build generative AI applications at scale. When associated with an agent, guardrails can help prevent unwanted behavior from model responses or user messages, stop your chatbot from responding to non-relevant queries, and protect personal information, ultimately improving the robustness of your agent implementation.  
**Sources**  
[Configure response generation for reasoning models and considerations - Amazon Bedrock](https://docs.aws.amazon.com/bedrock/latest/userguide/kb-test-configure-reasoning.html)  
[Improve LLM application robustness with Amazon Bedrock Guardrails and Amazon Bedrock Agents | AWS Machine Learning Blog](https://aws.amazon.com/blogs/machine-learning/improve-llm-application-robustness-with-amazon-bedrock-guardrails-and-amazon-bedrock-agents/)  
[Use cases for Amazon Bedrock Guardrails - Amazon Bedrock](https://docs.aws.amazon.com/bedrock/latest/userguide/guardrails-use.html)  
[Implement safeguards for your application by associating a guardrail with your agent - Amazon Bedrock](https://docs.aws.amazon.com/bedrock/latest/userguide/agents-guardrail.html)

Clarification on Bedrock Guardrails support for Claude 3.7 Sonnet with Bedrock Agents

Add your answer

Relevant content