Fully homomorphic encryption (FHE) is encryption that supports general computation directly on ciphertext, without revealing the underlying plaintext to the computing party. It produces an encrypted result that an authorized secret-key holder can decrypt. Its purpose is to keep inputs confidential during processing, not just while they are stored or transmitted.
How FHE Works
With fully homomorphic encryption, a client encrypts its inputs and gives an evaluator the ciphertext and a description of the computation. The evaluator performs supported operations without possessing the decryption secret, then returns an encrypted output. The authorized key holder decrypts that output to obtain the result corresponding to the intended plaintext computation, subject to the scheme's numerical precision.
Ciphertext operations correspond to arithmetic or logical operations on the underlying messages. Composing these operations makes more complex calculations possible. “Fully” means that arbitrary combinations and depths of operations are supported in principle, rather than only selected operations or a fixed shallow computation. This describes computational capability, not unlimited practical capacity: a calculation must still fit available time, memory and accuracy requirements.
Encrypted Cloud Processing
A cloud-processing example illustrates the value of fully homomorphic encryption. In a hypothetical workflow, a client encrypts a set of measurements and asks a provider to calculate their sum. The provider evaluates the sum over ciphertext and returns an encrypted total without receiving readable measurements. The client decrypts the total locally. Unlike ordinary cloud processing that decrypts data before calculation, this workflow keeps the inputs encrypted at the computing provider. It does not necessarily hide the requested computation or every observable detail, such as data sizes.
Looking for liquidity, exploring on-ramp/off-ramp services, or seeking expert guidance?
Get startedCosts and Numerical Limits
Fully homomorphic encryption generally requires more processing and storage than plaintext computation. Feasibility depends on the workload, chosen scheme, parameters and available hardware. Existing application code may need substantial changes to express calculations through supported encrypted operations. Larger ciphertexts, additional key material and intermediate results can also increase memory and data-transfer requirements.
In many schemes, operations increase ciphertext noise, which must remain controlled for reliable decryption. Bootstrapping refreshes ciphertext so further operations can proceed, but adds computational cost. Scheme-specific rescaling manages numerical scale in approximate arithmetic; it is not interchangeable with ciphertext refreshing. Approximate schemes also introduce numerical error, which must be assessed against the application's tolerance. Replacing a function with a polynomial can make encrypted evaluation practical, but introduces an approximation trade-off rather than universally improving accuracy or speed.
Security Boundaries and Key Management
The confidentiality benefit of fully homomorphic encryption does not cover every part of a system. FHE alone does not establish that an evaluator performed the requested computation correctly; verification requires additional mechanisms. Decrypted outputs may reveal sensitive facts about the inputs, even when encrypted evaluation was secure.
Compromised keys can enable unauthorized decryption, while compromised endpoints can expose data before encryption or after decryption. Secure key storage and controls on result access therefore remain necessary. Multiparty key arrangements require an appropriate protocol: FHE is not synonymous with secure multiparty computation. It does not inherently require a trusted third party, and appointing one is not itself a guarantee of integrity.