Data retention

Overview

Amazon Bedrock gives you explicit control over whether your prompts and outputs are retained from your inference requests. You can configure data retention at the account or project level, and the setting is enforced consistently across the Messages, Chat Completions, and Responses APIs.

Your data retention configuration is yours to manage. If your account or project is configured for zero data retention (data_retention_mode: none) and you invoke a model that requires retention, Amazon Bedrock will block the request and return an error — you always control your retention policy.

Data retention modes

Data retention is controlled by a mode rather than a simple on/off toggle:

How modes are ordered

Retention modes form an ordered scale, from least to most permissive:

none  <  default  <  aws_review  <  provider_data_share

A model is available to you when your effective mode is at or above the mode that model requires. A more permissive setting subsumes a less permissive one: if you have authorized AWS to share your content with the model provider, you have also authorized AWS to review it itself.

inherit is not part of this ordering — it expresses no opinion at its scope and defers to a broader one. See How your retention mode is determined.

How your retention mode is determined

Data retention is configured at two scopes, with the model's own default as the fallback:

The effective mode for any request is determined by taking the first scope whose value is not inherit:

effective mode = first non-inherit value of (project → account → model default)

For example, if your project is set to inherit and your account is set to none, the effective mode is none for all models invoked from that project.

Configuring data retention

Set account-wide data retention

curl -X PUT https://bedrock-mantle.us-east-1.api.aws/v1/data_retention \
  -H "x-api-key: $BEDROCK_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "mode": "aws_review" }'

Response:

{
  "mode": "aws_review",
  "updated_at": 1733529600
}

Bedrock Control Plane:

curl -X PUT https://bedrock.us-east-1.amazonaws.com/data-retention \
  -H "Authorization: Bearer $AWS_BEARER_TOKEN_BEDROCK" \
  -H "Content-Type: application/json" \
  -d '{ "mode": "aws_review" }'

Response:

{
  "mode": "aws_review",
  "updated_at": "2026-06-07T20:19:44.723Z"
}

Set project-level data retention

curl https://bedrock-mantle.us-east-1.api.aws/v1/organization/projects/proj_abc123 \
  -H "x-api-key: $BEDROCK_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "data_retention": { "mode": "aws_review" } }'

Check your current configuration

# Account level
curl https://bedrock-mantle.us-east-1.api.aws/v1/data_retention \
  -H "x-api-key: $BEDROCK_API_KEY"

# Project level
curl https://bedrock-mantle.us-east-1.api.aws/v1/organization/projects/proj_abc123 \
  -H "x-api-key: $BEDROCK_API_KEY"

Bedrock Control Plane:

# Account level
curl https://bedrock.us-east-1.amazonaws.com/data-retention \
  -H "Authorization: Bearer $AWS_BEARER_TOKEN_BEDROCK"

Check a model's effective mode and allowed modes

curl https://bedrock-mantle.us-east-1.api.aws/v1/models/anthropic.claude-fable-5 \
  -H "x-api-key: $BEDROCK_API_KEY"

Response:

{
  "id": "anthropic.claude-fable-5",
  "created": 1733443200,
  "owned_by": "system",
  "status": "available",
  "data_retention": {
    "mode": "aws_review",
    "source": "account",
    "allowed_modes": ["aws_review", "provider_data_share"]
  }
}

Model availability and data retention

Each model declares the retention modes that satisfy its requirement through allowed_modes, which lists every mode at or above the minimum the model needs. If your effective mode sits below what the model requires — see How modes are ordered — the model will appear as status: "unavailable" in the models list and requests to it will be blocked.

Example: Claude Fable 5 and Claude Fable 5.1 require human review (allowed_modes: ["aws_review", "provider_data_share"]). You must explicitly set your data retention mode to aws_review, or to the legacy provider_data_share, before you can invoke these models. If your effective mode is none or default, these models will be unavailable.

By setting aws_review, you are explicitly instructing us to retain your inputs and outputs so that AWS can perform the human review the model provider requires as a condition of access. Your content is not shared with the model provider. See Amazon Bedrock abuse detection and AWS Service Terms.

Response when a model is unavailable due to retention policy:

{
  "id": "anthropic.claude-fable-5",
  "created": 1733443200,
  "owned_by": "system",
  "status": "unavailable",
  "status_reason": "This model is not available under data retention mode 'default'.",
  "data_retention": {
    "mode": "default",
    "source": "account",
    "allowed_modes": ["aws_review", "provider_data_share"]
  }
}

How to opt in

To enable Claude Fable 5 and Claude Fable 5.1 for your account:

curl https://bedrock-mantle.us-east-1.api.aws/v1/data_retention \
  -H "x-api-key: $BEDROCK_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "mode": "aws_review" }'

Or at the project level (if you want to limit human review to a specific project):

curl https://bedrock-mantle.us-east-1.api.aws/v1/organization/projects/proj_abc123 \
  -H "x-api-key: $BEDROCK_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "data_retention": { "mode": "aws_review" } }'

Zero data retention (ZDR) access

Some models require data retention for safety and abuse-prevention purposes. If your organization requires zero data retention for compliance reasons and you need access to these models, contact your AWS account manager to discuss eligibility. ZDR access is evaluated on a per-account, per-model basis in coordination with the model provider.

Accounts approved for ZDR on a specific model will see "none" included in that model's allowed_modes.

Enforcing retention policy with IAM

You can enforce a data retention policy across your organization using IAM policies or Service Control Policies (SCPs). The write actions publish a bedrock-mantle:DataRetentionMode condition key that lets you restrict which modes can be set.

Example SCP — require zero data retention across the organization:

{
    "Effect": "Deny",
    "Action": [
        "bedrock-mantle:PutAccountDataRetention",
        "bedrock-mantle:CreateProject",
        "bedrock-mantle:UpdateProject"
    ],
    "Condition": {
        "StringNotEquals": {
            "bedrock-mantle:DataRetentionMode": "none"
        }
    }
}

Bedrock Control Plane:

{
    "Effect": "Deny",
    "Action": [
        "bedrock:PutAccountDataRetention"
    ],
    "Condition": {
        "StringNotEquals": {
            "bedrock:DataRetentionMode": "none"
        }
    }
}

This prevents anyone in the organization from setting data retention to anything other than none, ensuring no inference data is ever retained.

Example SCP — permit AWS retention but not human review:

{
    "Effect": "Deny",
    "Action": [
        "bedrock-mantle:PutAccountDataRetention",
        "bedrock-mantle:CreateProject",
        "bedrock-mantle:UpdateProject"
    ],
    "Condition": {
        "ForAnyValue:StringEquals": {
            "bedrock-mantle:DataRetentionMode": [
                "aws_review",
                "provider_data_share"
            ]
        }
    }
}

Use this when your organization accepts retention for abuse detection but cannot permit human review of its content. Models that require human review will appear as status: "unavailable" to accounts under this policy.

What data is retained and for how long

For models requiring aws_review (currently Claude Fable 5 and Claude Fable 5.1): user prompts and completions are retained within the AWS boundary for up to 30 days and may be reviewed by AWS to meet the human review requirement the model provider imposes as a condition of access. Your content is not shared with the model provider.

For the legacy provider_data_share mode: Amazon Bedrock does not share your content with model providers today, so this mode results in the same handling as aws_review — retained within the AWS boundary for up to 30 days, and reviewed by AWS only where the model requires it.

For models under default mode: data may be retained for abuse detection purposes — see Amazon Bedrock abuse detection for required retention details. For retention beyond abuse detection (e.g., Responses API with store=true), consult the model's documentation and terms.

If cross-region inference is enabled for these models, retained inputs and outputs are stored in destination regions (i.e., the region where your inference request is processed).

See Anthropic Terms of Service for model-specific data handling details.

IAM actions reference