Search

Retrieval Augmented Generation (RAG) systems have shown great promise in natural language processing. However, their reliance on data stored in a retrieval database, which may contain proprietary or sensitive information, introduces new privacy concerns. Specifically, an attacker may be able to infer whether a certain text passage appears in the retrieval database by observing the outputs of the RAG system, an attack known as a Membership Inference Attack (MIA). Despite the significance of this threat, MIAs against RAG systems have yet remained under-explored. This study addresses this gap by introducing an efficient and easy-to-use method for conducting MIA against RAG systems. We demonstrate the effectiveness of our attack using two benchmark datasets and multiple generative models, showing that the membership of a document in the retrieval database can be efficiently determined through the creation of an appropriate prompt in both black-box and gray-box settings. Moreover, we introduce an initial defense strategy based on adding instructions to the RAG template, which shows high effectiveness for some datasets and models. Our findings highlight the importance of implementing security countermeasures in deployed RAG systems and developing more advanced defenses to protect the privacy and security of retrieval databases., Comment: 16 pages, 3 figures

Regulated industries, such as Healthcare and Finance, are starting to move parts of their data and workloads to the public cloud. However, they are still reluctant to trust the public cloud with their most sensitive records, and hence leave them in their premises, leveraging the hybrid cloud architecture. We address the security and performance challenges of big data analytics using a hybrid cloud in a real-life use case from a hospital. In this use case, the hospital collects sensitive patient data and wants to run analytics on it in order to lower antibiotics resistance, a significant challenge in healthcare. We show that it is possible to run large-scale analytics on data that is securely stored in the public cloud encrypted using Apache Parquet Modular Encryption (PME), without significant performance losses even if the secret encryption keys are stored on-premises. PME is a standard mechanism for data encryption and key management, not specific to any public cloud, and therefore helps prevent vendor lock-in. It also provides privacy and integrity guarantees, and enables granular access control to the data. We also present an innovation in PME for lowering the performance hit incurred by calls to the Key Management Service. Our solution therefore enables protecting large amounts of sensitive data in hybrid clouds and still allows to efficiently gain valuable insights from it., Comment: 5 pages, 3 figures