Unified Guardrails for Amazon Bedrock: Cross-Account Safety Enforcement Now Generally Available

Introduction

Today, we announce the general availability of cross-account safeguards in Amazon Bedrock Guardrails, a capability that allows you to enforce and manage safety controls consistently across multiple AWS accounts within your organization. With this feature, you can define a single guardrail in your management account—via a new Amazon Bedrock policy—that automatically applies to all member accounts, organizational units (OUs), and generative AI applications. This centralized approach ensures uniform protection, reduces administrative overhead, and helps maintain compliance with responsible AI requirements.

Key Features of Cross-Account Safeguards

Organization-Level Enforcement

Organization-level enforcement applies a single guardrail from your management account to every entity within your AWS Organization. By configuring a policy, you can ensure that all Bedrock model invocations—across all member accounts and OUs—are automatically filtered by the same set of safeguards. This eliminates the need to manually replicate configurations per account and provides a baseline of protection that cannot be bypassed by individual teams.

Account-Level Enforcement

For more granular control, you can also set account-level enforcement. This allows you to apply a guardrail to all Bedrock inference calls within a specific AWS account. The configured filters are enforced on every API invocation, giving you the flexibility to add stricter controls for sensitive use cases while still benefiting from the overarching organizational policy.

Benefits of Centralized Management

With cross-account safeguards, your security and governance teams can establish a unified approach to AI safety. Instead of monitoring and verifying each account independently, you define policies once and enforce them everywhere. This reduces the administrative burden of configuration drift and ensures consistent adherence to corporate responsible AI standards. The solution also supports a mix of organization-wide and account-specific rules, so you can address unique application requirements without compromising overall safety.

Getting Started with Centralized Enforcement

To begin, you need a guardrail with a specific version—this ensures immutability and prevents member accounts from modifying the configuration. Complete the prerequisites, including setting up resource-based policies for guardrails. Then, navigate to the Amazon Bedrock Guardrails console.

1. Enable Account-Level Enforcement: Choose Create in the Account-level enforcement configurations section. Select the guardrail and version you want to apply automatically to all Bedrock inference calls from that account in the current Region.
2. Define Affected Models: Use the new Include or Exclude behavior to specify which models will be impacted by the enforcement.
3. Configure Content Controls: For system prompts and user prompts, choose between Comprehensive (apply guardrails to everything) or Selective (apply only to specific content types).

For organization-level enforcement, create a guardrail policy in the management account and assign it to your entire organization. The policy will automatically propagate to all member entities.

Customizing Enforcement: Model Selection and Content Control

The new enforcement capabilities include fine-grained controls to tailor the guardrails to your needs. You can define which models are covered—either by explicitly including certain models or excluding others. Additionally, content control for prompts can be set to Comprehensive (block all policy violations) or Selective (apply rules only to specific categories). This allows you to balance strict safety with operational flexibility.

Conclusion

Cross-account safeguards in Amazon Bedrock Guardrails mark a significant step toward simplified, scalable AI governance. By enabling centralized safety policies that span accounts, OUs, and applications, you can enforce consistent protection, reduce manual oversight, and accelerate the responsible deployment of generative AI. Start exploring the new capabilities in the console today.