AWS outlines data authorization for generative AI applications

Amazon Web Services (AWS) has published the first installment of a two-part blog series focusing on data security and authorization for generative AI applications. The post, updated to reflect the 2025 OWASP Top 10 for LLM entries, highlights the critical role of data authorization in business workload architectures.

The blog details the risks associated with using sensitive data – including customer, patient, supplier, and employee information, as well as intellectual property (IP), personally identifiable information (PII), or protected health information (PHI) – within generative AI workloads. It provides guidance on implementing data authorization mechanisms for applications such as fine-tuning foundation models (FM), retrieval augmented generation (RAG), and AI agents, including within Amazon Bedrock Agents.

AWS notes that generative AI introduces new methods for utilizing enterprise data. Unlike traditional AI solutions that often rely on internally labeled data, generative AI can combine private and public data, along with semi-structured or unstructured data from various sources.