Explore our new Cybersecurity Terms Glossary, explained by our experts.
Cybersecurity Terms Glossary
Homomorphic encryption allows computation on ciphertext. The result, once decrypted, matches what the same computation would have produced on the plaintext, which never has to be exposed.

What Is Homomorphic Encryption?
Conventional encryption forces a choice: either data stays protected, or it gets used. Anything that needs to search, total, compare, or model must first decrypt, which is why database servers, analytics platforms, and machine learning pipelines all end up holding plaintext. Homomorphic encryption breaks the trade-off. Operations performed on ciphertext produce a ciphertext that decrypts to the correct answer, and the party doing the computing never sees the inputs or the output.
Craig Gentry showed in 2009 that a fully homomorphic scheme was possible at all, using lattice-based constructions and a technique called bootstrapping to control the noise that accumulates with every operation. His first implementation took around thirty minutes per elementary operation. The field has improved by many orders of magnitude since, which is why it has started appearing in production.
The schemes you will meet in libraries are BFV and BGV for exact integer arithmetic, CKKS for approximate arithmetic on real numbers — the usual choice for machine learning, because it tolerates small errors — and TFHE for fast bootstrapping over boolean circuits. Microsoft SEAL, OpenFHE, HElib, and Zama’s Concrete are the practical implementations.
The technology is past the proof-of-concept stage in narrow, high-value cases:
| Dimension | Typical impact |
|---|---|
| Computation | Hundreds to many thousands of times slower than plaintext |
| Ciphertext size | Expansion of one to three orders of magnitude |
| Expressiveness | Branching on secret data is not possible; algorithms must be rewritten as circuits |
| Parameter choice | Security depends on noise budget and lattice parameters chosen per workload |
| Specialist skill | Scarce; parameter errors silently weaken security or corrupt results |
Comparisons, sorting, and conditional logic are the awkward cases, because the computing party cannot branch on a value it cannot read. Many designs that look natural on paper become impractical once written as a fixed-depth circuit.
Homomorphic encryption is one of several privacy-enhancing technologies, and rarely the cheapest. Secure multi-party computation splits a computation between parties so that none sees the whole input, and is often faster where the parties can interact. Trusted execution environments such as Intel SGX or AWS Nitro Enclaves keep data encrypted in memory and decrypt only inside the enclave, which is far faster but relies on hardware and has a history of side-channel research against it. Differential privacy addresses a different problem entirely: what the published results reveal, rather than who can read the inputs.
Importantly, homomorphic encryption does not solve access control. It protects data from the party performing the computation; whoever holds the decryption key still sees the result, and a flawed query can reveal more than intended. Under UK GDPR, encrypted personal data processed this way remains personal data to the controller, although the arrangement can materially reduce what a processor is exposed to — a point worth stating explicitly in a data protection impact assessment rather than assuming.
For most organisations the realistic posture today is to watch the field, identify the one or two use cases where sharing plaintext is genuinely blocking something valuable, and pilot there. Where privacy-preserving architecture is on your roadmap, our infrastructure and architecture review services can assess whether it earns its complexity.
For narrow, well-defined operations such as private lookups and aggregation, yes, and Apple and Microsoft have shipped features that rely on it. For general-purpose database or application workloads it remains too slow and too restrictive. Treat it as a specialist tool for specific problems rather than a replacement for conventional encryption.
Confidential computing decrypts data inside a hardware enclave, so performance is close to normal but you must trust the chip vendor and the enclave’s isolation. Homomorphic encryption never decrypts at all, so the trust assumption is purely mathematical, at a very large performance cost. The two are often combined rather than treated as rivals.
The main schemes are built on lattice problems, the same family underlying ML-KEM, and are currently believed to resist quantum attack. That is a useful property, but it is incidental: you would not choose homomorphic encryption for post-quantum protection when ML-KEM does that job at a tiny fraction of the cost.
That is precisely the promise, and it works for computations expressible as fixed-depth arithmetic circuits. The constraints are practical: the workload must be rewritten, the performance penalty must be acceptable, and your organisation still holds the keys and the resulting plaintext. It is an excellent fit for a narrow query, and a poor fit for a general application.
You May Be Interested In These
We use a few cookies that are needed to make this website work, and they are always on. We would also like to set analytics cookies, to see which pages are useful; functional cookies, so that our live chat works; and advertising cookies, to measure our campaigns. We will not set any of those until you say yes, and you can change your mind at any time from the Cookie Settings link in the footer.
Choose the cookies you are happy for us to use. Strictly necessary cookies are always on; everything else is off until you turn it on.
These cookies keep the website working: they hold your security checks, your form submissions and the choice you make here. The site cannot run without them, so they are always on.
These let our live chat open and remember your conversation, and let the newsletter sign-up work. Switch them off and the chat window will not appear.
These tell us which pages are read and how people find us, so that we can improve the site. The reports are about groups of visitors, never about you personally.
These let us measure our advertising and show you our adverts on other websites, such as LinkedIn, Google and Meta. Switch them off and you will still see adverts, but they will be less relevant.