The point: FHE enables ML models to compute on encrypted data without decrypting it, making cloud-based inferences on sensitive information (medical data, trade secrets) compliant with data protection regulations.
AWS demonstrates a method for machine learning inference with Amazon SageMaker in which data remains continuously encrypted from the time of request through the result—even from the cloud infrastructure itself. The solution leverages fully homomorphic encryption (FHE) and the concrete-ml library.
Fully homomorphic encryption (FHE) is an encryption technique that enables computations to be performed directly on encrypted data. In the context of ML inference, this means: a trained model can process an encrypted input and deliver an encrypted prediction without any intermediate values or requests being decrypted—not even for the SageMaker infrastructure itself.
AWS outlines concrete use cases: a health insurance company could provide cloud-hosted models for medical diagnosis without sharing patient data with third parties. In the energy sector, satellite imagery of potential drilling sites could be analyzed without disclosing sensitive geographic data. Telecommunications companies could process emails for spam and phishing detection while customer messages remain encrypted in compliance with regulations.
The practical approach uses concrete-ml, a high-level library specifically developed for FHE inference that is scikit-learn compatible and supports preconfigured model types. The workflow functions as follows: the model is trained with normalized data, deployed as an FHE version in SageMaker, and provisioned as a REST endpoint. Clients encrypt their requests, send them to the endpoint, receive encrypted predictions, and decrypt them locally.
This method complements (but differs from) confidential computing approaches such as AWS Nitro Enclaves, where data is processed in isolation but must be decrypted for computation. With FHE, the entire pipeline—request, processing, output—remains opaque to the cloud infrastructure.
Source: aws.amazon.com · Published June 8, 2026
Lumi AI News — AI-assisted curation pursuant to Article 50 EU AI Act. Paraphrase and classification by Lumi News Pipeline v1.6.5.