AWS ML BlogTuesday · June 9, 2026FREE

End-to-end encrypted ML inference with Amazon SageMaker AI and FHE

AWS published a blog post detailing how to perform end-to-end encrypted ML inference on Amazon SageMaker using fully homomorphic encryption (FHE) with the concrete-ml library. Unlike a previous post that used the low-level SEAL library to hand-craft a linear regression algorithm, this new approach leverages concrete-ml, a high-level library built specifically for FHE-based inference. concrete-ml supports several common model types out of the box and is API-compatible with scikit-learn, allowing developers to use familiar interfaces. The post demonstrates how to deploy a model that can make predictions on encrypted data without ever decrypting it, ensuring data privacy throughout the inference process. This capability is particularly relevant for industries handling sensitive information, such as healthcare or finance, where regulatory compliance requires data to remain encrypted even during processing. By integrating with SageMaker endpoints, the solution provides a scalable, managed infrastructure for secure real-time inferencing. The blog includes step-by-step instructions and code examples for setting up the encrypted inference pipeline, from training a model with scikit-learn to deploying it with concrete-ml on SageMaker.