Amazon Bedrock
AWS says Amazon Bedrock does not use customer content to train models or share that content with model providers to improve base models.
AWS states that customer inputs, outputs, and private model customisations are not used to train or improve the underlying base models.
No opt-out is required for the stated training baseline. Retention mode, model choice, invocation logging, and cross-region inference remain customer decisions.
AWS describes Bedrock content as isolated by customer and encrypted in transit and at rest in the AWS Region, subject to configured features and routing.
Current Bedrock documentation describes account- and project-level retention modes, including zero-retention controls and model-specific modes that may retain or share inference data.
The cited sources focus on service isolation, retention modes, and model-provider handling rather than one universal human-review statement.
Foundational reference pages for teams adopting, deploying, and operating production AI with inspectable controls.