Privacy-Preserving AI Inference via Fully Homomorphic Encryption

Lattica is a Tel Aviv-based startup founded in 2024 that offers a privacy-preserving inference platform for AI models. The company’s core technology uses fully homomorphic encryption (FHE) so that input data remains encrypted — even during processing by the model — thus ensuring that neither the AI provider nor any intermediary can access raw sensitive data.

Lattica implements a middleware layer named HEAL (Homomorphic Encryption Abstraction Layer), which is hardware-agnostic and designed to bridge encrypted data processing with standard ML infrastructure (GPUs, TPUs, CPUs, etc.). This architecture is intended to make encrypted inference compatible with existing AI pipelines without requiring organizations to redesign their systems.

The company emerged from stealth in April 2025, announcing a $3.25 million pre-seed funding round led by Cyber Fund, with participation from others including an angel investor associated with the Polygon Network.

Lattica targets sectors where data privacy is critical — for example, healthcare, finance, and regulated enterprises — offering them a pathway to deploy AI without exposing raw data. Its value proposition lies in enabling privacy-first AI by reconciling AI model utility with stringent confidentiality requirements, thus lowering one of the major barriers to AI adoption in sensitive domains.