The Privacy Tech Governments Won't Use: FHE, MPC and Zero-Knowledge, Explained

1. The false choice at the heart of every surveillance law

Every modern surveillance proposal rests on the same premise: to verify something about you, or to detect something harmful, a system must see your data. To check your age, it must read your birth date. To stop abuse material, it must inspect your messages. To prove your identity, it must hold your documents. The premise feels like common sense. It is also, increasingly, false.

A family of techniques collectively called privacy-enhancing technologies — PETs — breaks the link between 'verify' and 'see.' They let a system confirm a fact, run a computation, or flag a pattern without ever holding the underlying personal data in the clear. This is not science fiction or a startup pitch. The US National Institute of Standards and Technology runs a dedicated Privacy-Enhancing Cryptography project tracking exactly these methods, and NIST contributed to the 2025 National Privacy Research Strategy. The math is real, standardised, and improving fast.

2. Zero-knowledge proofs: prove it without showing it

A zero-knowledge proof lets you prove a statement is true while revealing nothing beyond its truth. You can prove you are over 18 without disclosing your birth date; prove you are a resident without revealing your address; prove you hold a valid licence without exposing the licence number. The verifier checks the proof mathematically and learns only the single bit they needed.

This is the technology behind the EU's age-verification 'mini-wallet,' and it is the cleanest example of the false choice being dissolved. The relying party gets its yes/no answer. The sensitive attribute never leaves your device. The catch is always implementation: a zero-knowledge system only protects you if the proofs are unlinkable and nothing phones home. The cryptography offers the guarantee; the deployment has to honour it.

3. Secure multi-party computation: compute on data nobody can see

Secure multi-party computation (MPC) lets several parties jointly compute a result over their combined inputs without any party revealing its input to the others. Hospitals can compute aggregate statistics across their patient populations without sharing patient records. Banks can detect fraud rings spanning multiple institutions without exposing individual customers. The output is computed; the inputs stay private.

NIST is actively maturing this area, including a call for multi-party threshold schemes (the IR 8214C effort) that overlaps the privacy-enhancing cryptography work. Threshold cryptography — splitting a secret or a signing key so no single party ever holds it whole — is a close cousin, and it is moving from research into standards. For identity systems, MPC and threshold techniques mean trust can be distributed rather than concentrated in one issuer who sees everything.

4. Homomorphic encryption: the holy grail, arriving slowly

Fully homomorphic encryption (FHE) allows arbitrary computation directly on encrypted data, without ever decrypting it. A server can run a search, a filter, or a machine-learning model over your encrypted information and return an encrypted result that only you can unlock. The server learns nothing — not the input, not the output. For a decade FHE was a beautiful idea that was far too slow to use; in the 2020s it became practical for narrow, valuable tasks, and NIST's threshold call explicitly invites FHE specifications with real implementations.

FHE matters for the surveillance debate because it removes the last excuse. Even where some computation genuinely must happen on a server — content classification, risk scoring — FHE shows it can, in principle, be done without the operator seeing the plaintext. The technology is not yet fast enough for every use case, but its existence reframes the question from 'can we avoid seeing the data?' to 'are we willing to pay for the version that does not?'

5. Differential privacy: the partner technology

The cryptographic PETs above protect data in use. Differential privacy protects the outputs. It adds carefully calibrated mathematical noise to published statistics so that the presence or absence of any single individual cannot be detected in the results — while the aggregate remains useful. It is how census bureaus and large platforms release data without leaking individuals. NIST frames differential privacy and privacy-enhancing cryptography as complementary: one guards the computation, the other guards what you reveal about it.

Together, these techniques form a complete toolkit. You can verify a fact (zero-knowledge), compute over private inputs (MPC), process encrypted data (FHE), and release safe statistics (differential privacy). At no point in that chain does anyone need to hold your raw personal data in the clear.

6. So why isn't this everywhere?

If the safe version exists, why do age-verification mandates, digital-ID schemes, and message-scanning proposals keep reaching for designs that see and store your data? Three honest reasons. Cost and speed: the privacy-preserving version is harder to build and sometimes slower to run, and 'ship it cheap' usually wins. Capability creep: a system that sees your data can be repurposed later for goals never disclosed at launch; an unlinkable, blind system cannot — which makes the surveilling version quietly more attractive to the people building it. Inertia: 'collect everything, secure it later' is the default pattern of two decades of software, and defaults are powerful.

None of these is a law of nature. They are choices. The role of an informed public — and of journalists and regulators — is to make the privacy-preserving choice the path of least resistance: to ask, every time a new verification system is proposed, 'you could build this with zero-knowledge proofs and never see my data. Why aren't you?' The technology has removed the excuse. What remains is whether we insist on it.

7. References