AIF-C01 Practice Question 30: Bedrock Customization Data Privacy

Amazon Bedrock uses the fine-tuning data you provide only to fine-tune a foundation model into a custom model for you, and for no other purpose. Your training data is not used to train the base models and is not distributed to third parties; other logged usage data is not used to train the models either. Custom models are encrypted at rest, by default with AWS owned keys, or with a customer managed AWS KMS key you control, and the training and validation data you supplied is not retained by Bedrock after the job completes (option B). A related point worth knowing: a fine-tuned model can replay some of its fine-tuning data in completions, so confidential values should still be filtered out of the training set.

- A is wrong because it describes the exact outcome Bedrock's terms rule out. Customization produces a separate model artifact for your account; the shared base model is untouched. - C is wrong because the data is not shared with the model provider at all. Bedrock deploys a copy of each provider's model into service-owned deployment accounts that the provider cannot access, so providers see neither prompts nor completions nor training data. - D is wrong because there is no opt-out to configure; reusing customer training data for base model training is not the default behaviour that an opt-out would protect you from.

Reference: Encryption of custom models in Amazon Bedrock