The Blind Machine: An End-to-End Platform for Privacy-Preserving Genomic Computation with Homomorphic Encryption

Many scientific questions require computing statistics across datasets that cannot be pooled because the underlying records contain sensitive information that cannot be shared. Homomorphic encryption enables computation directly on encrypted data without exposing plaintext, and more than a decade of research has demonstrated its practicality for privacy-preserving genomic analysis. Yet these systems are typically built one analysis at a time; no reusable, governed execution model has emerged that certifies and reuses approved encrypted computations independently of the specific study. To close this gap, we present The Blind Machine, an end-to-end platform for governed computation on encrypted data using homomorphic encryption and federated computing. The core design principle is that plaintext and secret keys never leave local machines, and the hosted service computes exclusively on ciphertext. Every experiment produces a machine-verifiable certificate, checkable offline, that binds the application, the committed encrypted inputs, the encrypted result, and the declared release policy. We show that the system is correct and practical through six curated biomedicalapplications built with the BFV scheme on seeded synthetic data, together with four studies on public IGSR/1000 Genomes genotypes. The work is fully reproducible: we release the open-source components, public-genome installer, synthetic data, experiment scripts, and an AI agent skill that helps reviewers reproduce the experiments.