Effective GenAI data security requires knowing:
- What sensitive data / IP / copyrighted information was used.
- If sensitive data was properly anonymized and de-identified.
- What data was used for fine tuning / model training / RAG (Retrieval Augmented Generation) databases.
- Any compliance violations that are being breached (including regulatory policies.
Bedrock ensures GenAI data controls through:
- Enhanced visibility: Bedrock discovers and categorizes GenAI data, assessing its materiality to the business. It utilizes fingerprinting to identify datasets involved in model training or RAG databases, and a Data Bill of Materials (DBOM) report provides a full understanding of data being used for GenAI model training / fine tuning and RAG.
- Smart detection: Bedrock establishes adaptive Trust Boundaries to monitor data and training usage outside established data perimeters. It alerts if IP / sensitive data or copyrighted information ends up in a RAG database or data for fine tuning.
- Automated response: Bedrock notifies on policy violations and can automatically remediate actions, such as preventing the ingestion of sensitive data into AI models.
- Reduced risk: Bedrock's workflows help prevent the unintentional indexing of sensitive data into RAGs or model training, minimizing duplication and controlling access permissions.
