Privacy-enhancing technologies: differential privacy and zero-knowledge proofs explained
Privacy-enhancing technologies (PETs) let an organisation get the answer it needs from personal data without holding or revealing more than that answer. Differential privacy adds measured random noise to published statistics, so nobody can tell whether a particular person is in the data. A zero-knowledge proof lets you prove something about yourself, such as being over 18, without handing over the details behind it. Both serve security as much as privacy: data that was never collected, or cannot be read, cannot be stolen.
Why less data is safer data
Every database of personal information is a target. Criminals break in to steal names, ID numbers and dates of birth because those details open bank accounts and take over identities. Even well-run organisations suffer breaches, so the safest record is often the one that was never kept.
South African law already points this way. Section 10 of the Protection of Personal Information Act (POPIA) says personal information may only be processed if it is adequate, relevant and not excessive for its purpose. Section 14 says records should not be kept longer than needed. The UK Information Commissioner's Office (ICO) describes PETs as one way to put those principles into practice: use less personal information, protect it better, and give people more control over it. The ICO is also clear that PETs are not a silver bullet; they support good data handling rather than replace it.
Differential privacy
Differential privacy is a mathematical way to publish useful statistics about a group while protecting each person in it. It works by adding small, carefully measured random changes to the results. Anyone looking at the output can learn about the group, but cannot work out with confidence whether any one person was included.
A simple way to picture it is the coin-flip survey, a textbook illustration. Ask people a sensitive yes-or-no question, but tell each of them to flip a coin in private first. Heads, they answer truthfully. Tails, they flip again and answer yes or no according to the second coin. No single answer proves anything about the person who gave it, yet across thousands of answers the true rate can still be estimated.
The amount of noise is set by a value called epsilon, which the ICO calls a privacy budget. A smaller epsilon means more noise and more privacy, at the cost of less precise results.
Where it is used
- The 2020 United States Census. The US Census Bureau explains that it took the published tables from the 2010 Census and used them alone to rebuild records for more than 300 million people, then matched those records to commercial data. Ordinary published statistics had leaked more than intended, so the Bureau based its 2020 disclosure protection on differential privacy.
- Apple devices. Apple's technical overview describes adding noise on the device before usage data for features such as emoji and keyboard suggestions leaves it, removing device identifiers and dropping IP addresses.
- Google's research and tools. Google's RAPPOR project (2014) used the same randomised-response idea to learn statistics from Chrome without identifying users, and Google publishes open-source differential privacy libraries.
The US National Institute of Standards and Technology (NIST) published final guidance on evaluating differential privacy, Special Publication 800-226, in March 2025.
Zero-knowledge proofs
A zero-knowledge proof lets one party prove to another that a statement is true without revealing anything beyond the fact that it is true. NIST describes it as proving knowledge of a secret without disclosing the secret.
The everyday example is age. Today an online shop that must check your age may ask you to upload a photo of your ID document, which shows your name, date of birth, address and ID number. The shop then holds a copy, and so does anyone who later breaks into the shop's systems. With a zero-knowledge proof, a digital ID wallet could tell the shop only "yes, this person is over 18". The shop learns what it needs and holds nothing worth stealing.
Where it is used
- Age checks in the European Union. The European Commission's age verification app is designed to let people prove they are over 18 without sharing any other personal information. Its technical specification includes a zero-knowledge proof scheme as an experimental feature.
- Privacy-focused digital currencies. Monero uses Bulletproofs, a form of zero-knowledge proof, to show that a transaction's hidden amounts are valid. Zcash uses a type of proof called a zk-SNARK for its shielded transactions.
- Standards work. ZKProof, an open industry and academic initiative, is working towards common standards, and NIST took part from 2019 to 2024.
Other privacy-enhancing technologies
| Technology | What it does | Plain example |
|---|---|---|
| Homomorphic encryption | Calculates on encrypted data without decrypting it | A service adds up encrypted figures and returns an encrypted total only you can open |
| Secure multiparty computation | Several parties compute a joint result without sharing their own inputs | Three companies learn their average spending without any of them seeing the others' figures |
| Federated learning | Trains a shared model on devices and combines only what each device learned | A phone keyboard improves its suggestions without sending what you type |
| Synthetic data | Generates artificial data with the same patterns as the real thing | Software is tested on realistic records that belong to no one |
The ICO lists honest limits for each. Homomorphic encryption can be thousands of times slower than ordinary processing. Federated learning updates can still leak information unless combined with other techniques. A zero-knowledge proof is only as good as its implementation.
What this means for a business
Most businesses will not build these techniques themselves. The principle behind them applies to everyone:
- Collect only what you need. If a date of birth is only used to check age, ask for a yes or no instead.
- Keep it only as long as you need it, as POPIA section 14 requires.
- Protect what you keep, with encryption at rest and in transit.
- Choose suppliers that do the same.
The technique most people can use today is end-to-end encryption. It is not one of the PETs on the ICO's list, but it follows the same idea: an email encrypted with OpenPGP can be read only by the people it was sent to, not by the servers it passes through. Our guide Why and how to encrypt your email with OpenPGP shows how.
For mail that never sits readable on a shared server, Noiz runs Private Email on a server for one organisation alone; ask us for a quote.
Sources
- NIST SP 800-226, Guidelines for Evaluating Differential Privacy Guarantees (March 2025): https://csrc.nist.gov/pubs/sp/800/226/final
- NIST, Privacy-Enhancing Cryptography project: https://csrc.nist.gov/projects/pec
- NIST, zero-knowledge proofs: https://csrc.nist.gov/projects/pec/zkproof
- ZKProof: https://zkproof.org/
- ICO, privacy-enhancing technologies: https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/data-sharing/privacy-enhancing-technologies/
- OECD, Emerging privacy-enhancing technologies (Digital Economy Papers No. 351, March 2023): https://www.oecd.org/content/dam/oecd/en/publications/reports/2023/03/emerging-privacy-enhancing-technologies_a6bdf3cb/bf121be4-en.pdf
- US Census Bureau, disclosure avoidance for the 2020 Census: https://www.census.gov/programs-surveys/decennial-census/disclosure-avoidance.2020.html
- US Census Bureau, Why the Census Bureau chose differential privacy (March 2023): https://www2.census.gov/library/publications/decennial/2020/census-briefs/c2020br-03.pdf
- Apple, Differential Privacy overview: https://www.apple.com/privacy/docs/Differential_Privacy_Overview.pdf
- Google Research, RAPPOR: https://research.google/pubs/rappor-randomized-aggregatable-privacy-preserving-ordinal-response/
- Google, differential privacy libraries: https://github.com/google/differential-privacy
- Google Research, federated learning (6 April 2017): https://research.google/blog/federated-learning-collaborative-machine-learning-without-centralized-training-data/
- European Commission, EU age verification: https://digital-strategy.ec.europa.eu/en/policies/eu-age-verification
- EU age verification specification, Annex B, zero-knowledge proofs: https://ageverification.dev/Technical%20Specification/annexes/annex-B/annex-B-zkp/
- Monero, Bulletproofs: https://www.getmonero.org/resources/moneropedia/bulletproofs.html
- Zcash, what are zk-SNARKs: https://z.cash/learn/what-are-zk-snarks/
- Protection of Personal Information Act 4 of 2013: https://www.gov.za/sites/default/files/gcis_document/201409/3706726-11act4of2013protectionofpersonalinforcorrect.pdf
Note: while some of this content was produced with the help of AI, the final editing has been done by a real human being.
