Encryption that lets a system compute on data while it stays encrypted, so the result is correct once decrypted and the data is never exposed.
Fully Homomorphic Encryption (FHE) allows AI models to process and run inference on encrypted data without decrypting it in memory. In principle, an enterprise can send sensitive data to a cloud-based LLM, receive an encrypted prediction, and decrypt the result locally, so the cloud provider cannot read the underlying data.
See also Air-Gapped Environment, Role-Based Access Control and Zero Trust Architecture.