OpenFHE

OpenFHE is the only open-source library providing a single, consistent C++ API across all four major FHE schemes:

• BFV: Exact integer arithmetic for financial computations
• BGV: Integer arithmetic with efficient modulus switching
• CKKS: Approximate real-number arithmetic for machine learning
• TFHE: Fast boolean gate bootstrapping for arbitrary circuits

This unification eliminates the need to learn separate libraries when switching between schemes for different stages of a privacy-preserving pipeline.

OpenFHE integrates hardware acceleration through multiple backends to dramatically reduce encrypted computation latency:

• Intel HEXL: Optimized for Intel AVX-512 instruction sets, accelerating number-theoretic transforms
• GPU acceleration: Offloads polynomial arithmetic to CUDA-capable GPUs
• ARM64 support: Enables FHE on edge and mobile devices

These backends are selectable at runtime without changing application code, allowing deployment across diverse hardware targets.

OpenFHE implements Threshold FHE and Multi-Key FHE protocols, enabling collaborative encrypted computation:

• Multiple parties encrypt data under their own distinct keys
• Joint computation proceeds without decrypting individual inputs
• Decryption requires a threshold of parties to collaborate

This is critical for federated learning and multi-institutional healthcare analytics where no single entity should access the combined plaintext.

Enables encrypted inference on sensitive patient data, such as medical imaging or genomic sequences, hosted on untrusted cloud infrastructure. A hospital can send an encrypted CT scan to a third-party AI diagnostic service, which processes it using a CKKS-based neural network and returns an encrypted diagnosis without ever seeing the raw image. This architecture satisfies HIPAA and GDPR requirements by ensuring the model host remains a zero-knowledge processor.

• Use Case: Encrypted radiology report generation
• Scheme: CKKS for approximate neural network inference
• Benefit: Model owner protects IP; hospital protects patient privacy

Allows multiple banks to collaboratively train or run fraud detection models on their combined transaction data without exposing individual customer records. Using threshold FHE, a model can compute risk scores over encrypted data pooled from several institutions, with decryption requiring a quorum of participants. This breaks down data silos that previously prevented effective detection of money laundering rings operating across multiple banks.

• Use Case: Cross-institutional anti-money laundering (AML)
• Scheme: BFV for exact integer arithmetic on transaction amounts
• Benefit: Joint analysis without violating banking secrecy laws

Facilitates secure computation over DNA sequences for personalized medicine and research. A patient's encrypted genome can be queried for specific genetic markers or used in a genome-wide association study (GWAS) without revealing the full sequence. OpenFHE's support for SIMD packing allows thousands of genetic variants to be processed in parallel, making population-scale encrypted analysis computationally feasible.

• Use Case: Secure querying of biobank data for clinical trial matching
• Scheme: BFV for exact matching on nucleotide sequences
• Benefit: Unlocks genetic data for research while preventing re-identification

Enables model providers to offer inference APIs without ever seeing user data, and users to get predictions without seeing the model weights. A proprietary transformer model can be evaluated entirely on encrypted input, producing encrypted logits. This dual-protection model is critical for commercializing AI where both the model IP and user data are sensitive assets. OpenFHE's hardware acceleration support via Intel HEXL makes this low-latency enough for interactive applications.

• Use Case: Secure inference for proprietary credit scoring models
• Scheme: CKKS with polynomial activation approximations
• Benefit: Simultaneous model IP protection and data privacy

Powers verifiable, private voting systems where individual ballots remain encrypted while the final tally is publicly computable. In a sealed-bid auction, all bids are submitted as ciphertexts; the auctioneer can homomorphically determine the winning bid and price without decrypting individual offers. OpenFHE's multi-key extensions allow each participant to encrypt under their own key, with decryption requiring collaboration, preventing any single party from unilaterally accessing sensitive intermediate results.

• Use Case: Sealed-bid spectrum auctions for telecom regulators
• Scheme: TFHE for fast boolean comparisons of bid values
• Benefit: Provably fair auctions with cryptographic privacy guarantees

Enables a messaging app to check a user's contact list against its user database to find friends already on the platform, without revealing the non-user contacts to the server. The client encrypts their address book; the server homomorphically computes the intersection and returns only the matching, still-encrypted identifiers. OpenFHE's implementation of labeled PSI extends this to associate metadata with matched entries, enabling richer discovery features without privacy leakage.

• Use Case: Privacy-preserving friend-finding in social apps
• Scheme: BFV for hashed contact identifier matching
• Benefit: Server learns nothing about non-users in the address book