# AWS guardrails means two things: Amazon Bedrock Guardrails for model text, and account-level controls such as SCPs

Facts read 16 Nov 2026. Page reviewed 16 Nov 2026.

AWS uses the word guardrails for two unrelated features. Amazon Bedrock Guardrails filters prompts and model responses. Account-level guardrails, meaning service control policies, resource control policies, declarative policies and Control Tower controls, cap what accounts can do. This page separates them, then covers the account kind, read 16 Nov 2026.

## Amazon Bedrock Guardrails filters what goes into and out of a model

AWS documentation says Amazon Bedrock Guardrails "provides configurable safeguards to help you build safe generative AI applications" and detects and filters undesirable content in user inputs and model responses. It sits at the application layer, next to a model call, and does not control which AWS accounts, Regions or services a person can use.

The documentation lists these safeguards: content filters (Hate, Insults, Sexual, Violence, Misconduct and Prompt Attack), denied topics, word filters, sensitive information filters that block or mask personal data, contextual grounding checks, and Automated Reasoning checks. A guardrail has a working draft and a built-in test window. After testing, the owner creates a version and passes the guardrail ID and version on an inference call, or calls the ApplyGuardrail API without invoking a model (AWS documentation, read 16 Nov 2026).

Everything below is about the other meaning. A search for model input filtering belongs with the Bedrock documentation, and this site does not cover it further.

## Account-level AWS guardrails are policies that cap what accounts can do

AWS Organizations documentation describes a service control policy as defining "a permission guardrail, or sets limits" on what IAM users and roles can do. The library on this site records 40 AWS entries across four mechanisms. Each entry states what it blocks and what it breaks.

| Mechanism | What it does | Library entries |
| --- | --- | --- |
| [Service control policy](/mechanisms/scp/) | Limits the actions available to principals in member accounts | 27 |
| [Resource control policy](/mechanisms/rcp/) | Limits access to resources in member accounts, whoever the caller is | 6 |
| [Declarative policy](/mechanisms/declarative-policy/) | Sets a service configuration baseline in the service control plane | 5 |
| [Permissions boundary](/mechanisms/permissions-boundary/) | Caps what an IAM user or role can be granted | 2 |

## Control Tower controls bundle those mechanisms under one console

AWS Control Tower implements preventive controls with service control policies and resource control policies, detective controls with AWS Config rules, and proactive controls with CloudFormation hooks. AWS says it is moving from the word guardrail to the word control and treats the two as synonyms. [AWS Control Tower guardrails](/aws-control-tower-guardrails/) covers the three types and their limits.

The two feature families differ in who writes the rule. In Control Tower the rule comes from an AWS-owned catalog. With a plain service control policy the rule is a JSON document you write, such as [Deny member accounts leaving the organization](/guardrails/aws-scp-deny-leave-org/).

## One library entry connects both meanings: a declarative policy that enforces a Bedrock guardrail

AWS documents a declarative policy that references a versioned Bedrock guardrail from the management account and applies it to inference calls in a Region. The library records the conditions: the guardrail must be owned by the management account, must have a non-DRAFT version, and needs a resource-based policy that lets organization members call ApplyGuardrail. See [Enforce an Amazon Bedrock guardrail on model inference](/guardrails/aws-declarative-bedrock-guardrail-enforcement/).

A way to test that policy before it ships: none found on 10 Nov 2026 in the AWS pages read. For the general method, see [testing a guardrail before you apply it](/guides/testing-a-guardrail-before-you-apply-it/).

## The quickest test for which meaning you need is the object being restricted

Account-level guardrails are the subject of this site. The [library](/library/) lists all 100 entries for AWS, Azure and Google Cloud, and the [method](/method/) explains how platforms that manage them are ranked.

- If the thing restricted is a prompt or a model response, you need Bedrock Guardrails.
- If the thing restricted is an API action, a Region, an encryption setting or a resource policy in an AWS account, you need an account-level guardrail.
- If the thing restricted is a CloudFormation stack before it deploys, you need a Control Tower proactive control or an engine such as [CloudFormation Guard](/engines/cloudformation-guard/).

## Related pages

- [AWS Control Tower guardrails: now called controls](/aws-control-tower-guardrails/)
- [Multi cloud governance: three hierarchies, three mechanisms](/multi-cloud-governance/)
- [What is cloud governance? The security case](/what-is-cloud-governance/)
- [All of them](/guides/)

## Sources

- AWS: Detect and filter harmful content by using Amazon Bedrock Guardrails: https://docs.aws.amazon.com/bedrock/latest/userguide/guardrails.html (read 16 Nov 2026)
- AWS Organizations: Amazon Bedrock policy syntax: https://docs.aws.amazon.com/organizations/latest/userguide/orgs_manage_policies_bedrock_syntax.html (read 10 Nov 2026)
- AWS Organizations: service control policies: https://docs.aws.amazon.com/organizations/latest/userguide/orgs_manage_policies_scps.html (read 16 Nov 2026)
- AWS Control Tower: how controls work: https://docs.aws.amazon.com/controltower/latest/userguide/how-controls-work.html (read 10 Nov 2026)

Last reviewed 16 Nov 2026. Canonical: https://cloudgovernancetools.com/aws-guardrails/
