A blind signature lets you get a message signed without showing the signer what they are signing. The signer does not learn the message, the recipient walks away with a normal-looking signature on a known message, and a verifier with the public key cannot tell that anything blind happened. The flagship use case is anonymous tokens and electronic cash: a bank issues digital coins by blindly signing them, so customers can spend the coins later without the bank linking the spending back to the issuance.
Blind signatures have been around since David Chaum invented them in 1982, and most deployed schemes still use RSA or Schnorr. Both fall to a sufficiently large quantum computer. The research question for the post-quantum era: can we build a blind signature on lattices, codes, or hashes that is efficient enough for production? This article explains what blind signatures do, why making them post-quantum is harder than making regular signatures post-quantum, and where the field stands.
What Blind Signatures Actually Do
Imagine you walk into a bar and need an "I am over 21" stamp on your hand. The bouncer looks at your ID, but you do not want them to know which bar you go to next. So you put a piece of carbon paper over your hand, the bouncer stamps the paper, and you keep the imprint on your skin. The bouncer signed something, but never saw what they signed.
In digital form, this is the blind signature protocol:
- The user picks a message m, generates a random "blinding factor" r, and sends the signer a blinded message m' = blind(m, r).
- The signer signs m' using their private key, producing a blinded signature s'.
- The user "unblinds" s' using r, recovering a normal signature s on m.
- Anyone with the signer's public key can verify s on m.
The signer learned nothing about m. The verifier sees a normal signature on m. The user has perfect plausible deniability about which messages they got signed. The math has to make this work out, and the standard constructions (RSA blind, Schnorr blind, BLS blind) all use group operations that quantum computers break.
Use cases:
- Anonymous tokens: Cloudflare's Privacy Pass uses blind signatures so users can prove humanness without being trackable.
- E-cash: Chaum's original construction was for anonymous digital cash. GNU Taler is a modern descendant.
- Voting: A registrar blindly signs ballots so the count is correct but the voter is anonymous.
- Anonymous credentials: A government issues blind-signed credentials, the citizen presents them at services without linking the issuance to the use.
- Privacy-preserving analytics: Apple's "Private Click Measurement" uses blind signatures to count ad conversions without identifying users.
Why Blind Is Harder Than Plain Signatures
A normal signature scheme has security against forgery: an attacker without the private key cannot produce a valid signature. A blind signature scheme has two security properties:
- One-more unforgeability: After getting k blind signatures, an attacker cannot produce a (k+1)th signature. This is harder than ordinary unforgeability because the attacker has chosen the messages.
- Blindness: The signer cannot link the blinded protocol transcript to the unblinded signature. Even if the signer keeps records of every signing session, they cannot tell which output corresponds to which input.
Both properties together are tricky to achieve. The classic Schnorr-style 3-round blind signature was thought secure for years until ROS attacks showed it could be broken when an attacker runs many parallel signing sessions. Modern blind signatures fix this with extra rounds or with "concurrent security" proofs.
For lattice-based blind signatures, you have to satisfy both unforgeability and blindness against quantum adversaries, while maintaining acceptable signature sizes. This is harder than plain ML-DSA, which only needs unforgeability. See ML-DSA vs SLH-DSA for the standard signatures, blind variants are an extension.
The State of Lattice Blind Signatures
Several lattice-based blind signature schemes have been proposed, but none is standardised and most have efficiency caveats. Notable constructions:
Ruckert (2010): One of the first lattice blind signatures, based on hash-and-sign over NTRU lattices. Signature sizes around 50 KB, slow signing. Mostly historical interest.
Hauck-Kiltz-Loss (HKL) blind signatures: A 3-round protocol that works over Module-LWE, similar to the lattice family that ML-DSA uses. Signatures are several KB. The security proof is in the random oracle model and concurrent security has been a focus of follow-up work.
Agrawal-Stehle-Yadav blind signatures: A 2-round construction with smaller signatures than HKL but more complex security analysis. Their paper (2022) reduces blind signing to standard lattice assumptions.
del Pino-Katsumata-Maller-Mouhartem-Plancon-Roux: A more recent blind signature based on Lyubashevsky-style sigma protocols, with concurrent security proofs that survive ROS-like attacks.
Concurrent attacks and defences: Benhamouda-Lepoint-Loss-Orru-Raykova showed that several previous lattice blind signatures had subtle vulnerabilities when run in concurrent sessions. The fixes involve longer protocols or different randomness handling. This is an active area as of 2026.
For the underlying math, see Lattice-Based Cryptography Explained.
Hash-Based and Code-Based Blind Signatures
Lattices are not the only candidate.
Hash-based blind signatures: Naive hash-based signatures (like SLH-DSA) do not blind cleanly, because the signature includes a Merkle authentication path that is hard to randomise. Some research proposes blind variants using oblivious hashing techniques, but signatures are typically over 100 KB.
Code-based blind signatures: Code-based schemes have a few proposals but most have very large signatures (multiple megabytes). The structure of code-based decoding does not split blindly as easily as lattice operations.
Symmetric blind signatures: A different approach uses MPC over symmetric primitives (AES, SHA-3) to simulate a blind signature. The result is post-quantum (since AES-256 needs only Grover speedup which can be defeated by doubling key length) but bandwidth-heavy.
For practical blind signatures, the lattice path is the most active.
What Privacy Pass Does Today
Cloudflare's Privacy Pass (used by hCaptcha, Apple, and others) is the most prominent production blind-signature deployment. The current production protocol uses VOPRF (verifiable oblivious pseudorandom functions) over elliptic curves, specifically P-384 or Ristretto255. This is not post-quantum.
The Privacy Pass IETF working group has discussed PQ migration. Two paths are on the table:
- Hybrid mode: Wrap the existing VOPRF in a lattice-based blind signature, similar to how TLS hybrid handshakes work.
- Pure PQ replacement: Replace VOPRF entirely with a lattice-based VOPRF. Several research proposals exist (Albrecht-Davidson-Deo-Smart 2021), but no standard yet.
Apple's Private Access Tokens, built on Privacy Pass, has a similar migration path. As of 2026, no production blind-signature system has migrated to post-quantum primitives.
ROS Attacks and Concurrent Security
A subtle issue that has affected many blind signature schemes is the ROS (Random Inhomogeneous Linear Equations Over a Finite Set) attack. The attack lets an adversary who can run many parallel signing sessions forge signatures on chosen messages. ROS attacks were known against blind Schnorr and several variants for years before they became practical, and Benhamouda-Lepoint-Loss-Orru-Raykova (2021) showed efficient ROS attacks against widely-deployed blind Schnorr.
For lattice blind signatures, similar concurrent-session attacks can apply. The Lyubashevsky-style sigma protocols that underlie most lattice signatures have analogues to the ROS structure, and several earlier lattice blind signatures had concurrency vulnerabilities. The fixes involve:
- Round-optimal protocols: 2-round constructions with statistical hiding properties.
- Commitment-based blinding: Adding a commitment phase that prevents the attacker from running too many parallel sessions.
- Tight concurrent security proofs: Modern lattice blind signature constructions explicitly prove security in concurrent sessions.
For production-grade blind signatures, concurrent security is non-negotiable. Schemes without it can be broken by an attacker who controls the order of signing requests. The research literature is converging on schemes that are concurrent-secure by design, but library maturity has not caught up yet.
For the broader context of lattice signatures, see ML-DSA vs SLH-DSA.
Performance Reality
Concrete benchmarks for current research-grade lattice blind signatures:
- HKL blind signature: Signing time 30 to 100 milliseconds, signature size 10 to 20 KB.
- del Pino et al. blind signature: Signing time 10 to 50 milliseconds, signature size 5 to 15 KB.
- Agrawal-Stehle-Yadav: Signing time 50 to 200 milliseconds, signature size 8 to 25 KB.
Compare to classical blind Schnorr: 1 to 5 milliseconds, 64 byte signatures. The size penalty is 100x to 400x. The time penalty is 10x to 100x. For low-volume use cases (research analytics, niche credential issuance), these costs are tolerable. For internet-scale deployments (Privacy Pass token issuance), they are still painful.
What QNSQY Does and Does Not Cover
QNSQY focuses on file encryption and signing using NIST-standardised primitives ML-KEM and ML-DSA. It does not include blind signatures, anonymous tokens, or privacy-pass-style protocols. The signing operations in QNSQY are normal ML-DSA where the signer knows what they sign.
For users who need anonymous tokens or e-cash, the recommendation is:
- Use a Privacy Pass implementation if classical blind signatures are acceptable for now.
- Watch the lattice blind signature literature for production-ready libraries.
- Plan a migration to a PQ blind signature when one matures past the research stage.
The two layers do not overlap: QNSQY protects the data, blind signatures protect the issuance metadata.
Threshold and Distributed Variants of Blind Signatures
Beyond simple two-party blind signatures (signer plus user), more elaborate variants exist. Threshold blind signatures split the signer's role across multiple parties such that no single party can sign alone, and the resulting signature is still blind to all of them. This is useful for systems where the signing authority should not be trusted as a single point of compromise.
Lattice threshold blind signatures are an active research area as of 2026. The main proposals build on Agrawal-Stehle-Yadav and del Pino et al. with additional structure for distributing the signing key. The challenges include keeping signatures small (already a problem in non-threshold lattice blind signatures), proving security in the presence of malicious participants, and managing the round complexity. Practical deployments are at least several years away.
For e-cash applications, threshold blind signatures are particularly attractive because they let the issuance authority be split across regulators, central banks, and commercial banks rather than concentrated in a single entity. GNU Taler's roadmap includes threshold-style enhancements over time. The IETF blind signature working group has discussed but not yet standardized threshold variants.
For applications that need blind signatures plus distributed signing today, the practical approach is to combine a non-threshold classical blind signature (RSA blind or VOPRF) with classical threshold cryptography (threshold ECDSA or threshold Schnorr) and migrate both layers to lattice-based variants as the research matures. This compounds the migration complexity but is the only path that preserves both properties simultaneously.
Frequently Asked Questions
Is any blind signature scheme post-quantum and standardised?
Not yet. The IETF and NIST have not standardised any post-quantum blind signature. Several lattice schemes (HKL, Agrawal-Stehle-Yadav, del Pino et al.) are well-studied research candidates, but none has gone through formal standardisation. Production deployments (Privacy Pass, Apple Private Access Tokens) still use classical VOPRF over elliptic curves.
Why does Privacy Pass not use ML-DSA directly?
ML-DSA is a normal signature scheme: the signer sees the message. Privacy Pass needs the signer to not see the message, which requires a blind signature scheme. ML-DSA does not have a standardised blind variant. Building a blind signature on top of ML-DSA's structure is a research direction but not finished.
Can I just use ML-DSA and trust the signer not to log?
That is one approach, but it removes the cryptographic guarantee that the signer cannot link signing to use. The whole point of blind signatures is that the user does not have to trust the signer about anonymity. If you are willing to trust the signer, regular ML-DSA works.
How big are lattice blind signatures?
Current research schemes produce signatures from a few KB to tens of KB, depending on the construction and security level. This is much larger than the 64 to 96 byte signatures of classical blind Schnorr or BLS. For some applications (privacy-pass-style anonymous tokens) the size is acceptable. For others (e-cash with high-volume circulation), it is still too large for production at internet scale.
Are there hash-based blind signatures?
Some research proposals exist, often using MPC techniques to simulate the blinding. Signatures are typically over 100 KB and signing is slow. Not currently practical.
How does blind signature research relate to zero-knowledge proofs?
Blind signatures and zero-knowledge proofs are related but distinct primitives. Both let you prove statements without revealing underlying data. Zero-knowledge proofs are more general (prove any computable predicate); blind signatures are specific (get a signature on hidden data). Many blind signature constructions internally use zero-knowledge components, particularly for proving that the blinded message is well-formed. Modern research often blurs the line, with constructions like blind ZK-SNARKs exploring overlap. Both areas face the same post-quantum migration challenge: classical primitives based on discrete log fall to Shor's algorithm, and lattice replacements have larger sizes and more complex security proofs.
What is the timeline for production-ready post-quantum blind signatures?
Best estimates from the research community suggest 2027 to 2030 for the first standardised post-quantum blind signature, followed by 2 to 3 years for library maturity and another 2 to 3 years for deployment in major systems like Privacy Pass. This puts production-grade post-quantum blind signatures at 2030 to 2035. Until then, applications that need blind signatures use classical schemes and accept the harvest-now-decrypt-later exposure for the issuance metadata.
Sources
- Chaum, D. "Blind Signatures for Untraceable Payments." CRYPTO 1982. (No online copy due to era; see Chaum's collected works.)
- Hauck, E., Kiltz, E., Loss, J. "A Modular Treatment of Blind Signatures from Identification Schemes." EUROCRYPT 2019. https://eprint.iacr.org/2018/1052
- Agrawal, S., Stehle, D., Yadav, A. "Towards Practical and Round-Optimal Lattice-Based Threshold and Blind Signatures." CCS 2022. https://eprint.iacr.org/2021/381
- del Pino, R., Katsumata, S., Maller, M., Mouhartem, F., Plancon, M., Roux, A. "Lattice-Based Blind Signatures, Revisited." CRYPTO 2024. https://eprint.iacr.org/2023/192
- Davidson, A., et al. "Privacy Pass: Bypassing Internet Challenges Anonymously." PoPETs 2018. https://eprint.iacr.org/2017/783
- Albrecht, M.R., Davidson, A., Deo, A., Smart, N.P. "Round-Optimal Verifiable Oblivious Pseudorandom Functions from Ideal Lattices." PKC 2021. https://eprint.iacr.org/2019/1271
- IETF Privacy Pass Working Group. "Privacy Pass Architecture and Specifications." https://datatracker.ietf.org/wg/privacypass/about/
Related Articles
- ML-DSA vs SLH-DSA
- Lattice-Based Cryptography Explained
- What Is Post-Quantum Cryptography?
- Hybrid Encryption
- NIST FIPS Guide for Post-Quantum Standards
- Code-Based Post-Quantum Cryptography: Future Beyond HQC
- Post-Quantum Attribute-Based Encryption: Research Status
- Concat KDF (SP 800-56C): The Two-Step Approach
- Post-Quantum Zero-Knowledge Proofs Overview
Protect Your Data Before Q-Day Arrives
QNSQY's NIST-standardized post-quantum encryption protects files against both current and quantum-era threats.