Homomorphic Encryption

Homomorphic encryption is a form of encryption that permits computation on ciphertext, producing encrypted results that, when decrypted, match the output of operations performed on the plaintext—enabling data processing without ever exposing the underlying data.

Definition

Homomorphic encryption (HE) is a class of encryption schemes that allows mathematical operations to be performed directly on ciphertext without decrypting it first. The result, once decrypted, is identical to the result of performing those same operations on the unencrypted plaintext. In formal terms: if Enc(a) + Enc(b) = Enc(a + b), the encryption is additively homomorphic; if Enc(a) * Enc(b) = Enc(a * b), it is multiplicatively homomorphic. A scheme supporting arbitrary combinations of both is fully homomorphic encryption (FHE).

Craig Gentry published the first viable FHE construction in 2009, using ideal lattice-based cryptography. Before that, mathematicians had theorized about the possibility for over 30 years without a working scheme. Gentry’s breakthrough proved that encrypted computation was not just theoretically possible—it was constructible.

Why It Matters

The global homomorphic encryption market reached $256 million in 2024 and is projected to exceed $550 million by 2028, according to Allied Market Research. The acceleration is driven by a structural conflict: organizations need to process data they are no longer permitted to see.

Under GDPR, CCPA, and emerging frameworks like Switzerland’s revised Federal Act on Data Protection (revFADP), processing personal data requires a legal basis—and that basis is increasingly difficult to establish when data crosses organizational or jurisdictional boundaries. FHE offers a radical alternative: process the data without ever accessing it. The data remains encrypted throughout the entire computation lifecycle. The processor sees ciphertext. The processor returns ciphertext. Only the data owner holds the decryption key.

For AI workloads, this means a model could theoretically process an encrypted prompt and return an encrypted response—without the model operator ever seeing the input or output in plaintext.

How It Works

Homomorphic encryption schemes are built on computationally hard mathematical problems, typically lattice-based:

Key generation: The data owner generates a public/private key pair. The public key encrypts data in a way that preserves algebraic structure. The private key decrypts the results.
Encryption: Plaintext values are encrypted using the public key. The ciphertext is significantly larger than the original data—current FHE schemes produce ciphertext 1,000x to 1,000,000x larger than the plaintext.
Evaluation: A third party performs computations (addition, multiplication, comparison) directly on the ciphertext. Each operation introduces noise into the ciphertext. After enough operations, the noise overwhelms the signal and the result becomes undecryptable.
Bootstrapping: Gentry’s key insight was a technique called bootstrapping—homomorphically evaluating the decryption function itself to reset the noise level. This makes arbitrary-depth computation possible, at significant computational cost.
Decryption: The data owner decrypts the final ciphertext with their private key, yielding the correct computation result.

The performance overhead remains the primary barrier. A 2024 benchmark by Intel’s HE Toolkit showed FHE operations running 10,000x to 100,000x slower than equivalent plaintext operations, depending on the scheme and circuit depth. Libraries like Microsoft SEAL, OpenFHE, and Zama’s TFHE-rs are narrowing this gap, but FHE is not yet viable for real-time interactive applications.

Stealth Cloud Relevance

Stealth Cloud does not deploy fully homomorphic encryption in its Phase 1 architecture—the performance overhead makes it incompatible with the sub-200ms latency requirement for Ghost Chat. Instead, Stealth Cloud achieves comparable privacy guarantees through a layered approach: client-side PII stripping, end-to-end encryption via the Web Crypto API, processing in ephemeral V8 isolates, and cryptographic shredding on session termination.

FHE represents the theoretical endgame for Stealth Cloud’s roadmap. When FHE performance reaches practical thresholds for LLM inference—a milestone multiple research groups are targeting before 2030—it will enable AI processing where even the ephemeral decryption step inside the isolate is eliminated. The server would compute on ciphertext end to end. Until then, zero-knowledge architecture combined with ephemeral infrastructure provides the strongest practical guarantee available.

The Stealth Cloud Perspective

Homomorphic encryption is the mathematical proof that computation does not require knowledge—that a machine can transform data it cannot read. Stealth Cloud treats this as a design north star: every architectural decision moves toward the day when the server processes everything and knows nothing.