Skip to content
INTERNET INFRASTRUCTURE & NETWORKS

Preparing the Web for Quantum Threats: The Evolution of Post-Quantum Privacy Pass and MOLE

As the technology industry looks ahead to the eventual arrival of cryptographically relevant quantum computers, a massive transition is already underway across the digital landscape. While the highest priorities for post-quantum cryptography focus on foundational building blocks like authenticated key-agreement protocols and Transport Layer Security (TLS), experts warn that modern internet security relies on quantum-vulnerable cryptography for a much wider array of services. Among these critical technologies are privacy-preserving tools used daily by site operators to protect applications from abuse while balancing user confidentiality.

Today, protocols like Privacy Pass are deployed extensively by major internet infrastructure providers, including Cloudflare, to manage edge traffic and safeguard websites against automated abuse. A primary function of these systems is individual client rate limiting, ensuring that no single entity or automated script can consume an unfair share of a site’s server resources. Historically, maintaining this balance meant forcing users to endure friction or give up personal data through mechanisms like CAPTCHAs, persistent accounts, or invasive browser fingerprinting. Privacy Pass was specifically engineered to dismantle this trade-off between user privacy and traffic moderation.

The core mechanism relies on issuing clients a batch of unlinkable cryptographic tokens that can later be redeemed to gain seamless access to a site. Because each token cannot be linked back to the specific client it was originally issued to, users can navigate the web privately without changing their everyday behavior. Over a decade of real-world deployment experience—utilizing Privacy Pass for privacy proxies like iCloud Private Relay and routing traffic through specialized gateways—has proven the viability of this approach. Now, industry collaborators and browser vendors are working together to bring the next generation of this technology to the web.

These upcoming frameworks aim to achieve two major objectives. The first is the development of sophisticated tools capable of managing agentic traffic, which threatens to exhaust server resources and degrade performance for human visitors if left unchecked. The second goal is fostering a more open ecosystem by lowering barriers to entry so that a wider variety of client applications can easily participate. Achieving these ambitious milestones, however, requires creating advanced capabilities that go far beyond what traditional Privacy Pass architectures currently offer.

An Unexpected Advantage in Post-Quantum Readiness

Given the trajectory of quantum computing research, engineers are actively evaluating the prospects of securing private authentication against future threats. Curiously, the ecosystem may currently be in a better structural position than many realize. While Privacy Pass continues to rely on classical cryptographic algorithms such as RSA and elliptic curves—meaning a sufficiently powerful quantum adversary could theoretically forge tokens at will and bypass rate-limiting measures—the protocol remains unconditionally private. No attacker, quantum or otherwise, can link the redemption of a token to its original issuance.

Furthermore, the advanced capabilities currently under development for the next generation of private authentication technology are being engineered to uphold this exact guarantee, ensuring that user interactions remain private even after the arrival of quantum computing. While this inherent privacy buys the ecosystem some valuable breathing room, the need for robust alternatives is rapidly approaching.

Much like the broader transition to post-quantum TLS, the primary hurdle facing researchers is the sheer physical size of post-quantum cryptographic primitives. For example, ML-DSA-44 digital signatures require a hefty 2,420 bytes, whereas traditional ECDSA-P256 signatures take up a mere 64 bytes. Because drop-in replacements are unlikely to materialize anytime soon, engineers will be forced to adopt primitives with higher performance costs and completely re-engineer underlying systems to bear that weight. A potential roadmap for post-quantum web privacy is now emerging, driven by a recent breakthrough in the design of blind signatures.

What We Need to Build: A Brief Look at MOLE

To understand where post-quantum privacy is heading, it is helpful to examine an emerging technology framework currently known as MOLE, an acronym for "Moderation Of unLinkable Endorsements." At the heart of traditional Privacy Pass lies a blind signature scheme, which allows a client to obtain a cryptographic signature under an issuer’s secret key without revealing the underlying message to the issuer. The client can subsequently present the message and the blind signature later on, maintaining a complete separation between the issuance and redemption phases.

In a standard Privacy Pass configuration, a token consists of a message combined with a blindly issued signature. To verify the token, a verifier must check it against the issuer’s public key, inherently learning the identity of the issuer in the process. While this effectively protects the identity of the client, it does not protect the identity of the issuer. For certain commercial use cases, this is an intentional design choice; knowing the issuer provides valuable signal regarding the type of client connecting to an edge network, such as verifying that a user is authorized to utilize a specific virtual private network product.

However, for many common web applications, revealing the issuer’s identity provides far more information than is necessary to grant access to a resource. Normally, a site operator only needs to know that a token was issued by some trusted entity rather than a specific one. MOLE takes this privacy concept a step further by entirely hiding the identity of the issuer, ensuring that verifiers cannot treat tokens differently based on who originally issued them. This guarantees equal treatment for all participants regardless of their underlying client software.

This advanced issuer-hiding property is achieved through a new cryptographic object called an endorsement, which is issued by an entity known as an anchor and redeemed by a moderator. Before redemption takes place, the client and the moderator mutually agree upon an anchor set consisting of the public keys of all anchors trusted by the moderator. During the redemption phase, the client proves that it received an endorsement from at least one of the anchors within that designated anchor set, without revealing which specific anchor performed the signing. Upon successful acceptance, the moderator issues the client a MOLE credential used to moderate subsequent interactions with the target site.

While endorsements must be anchor-hiding, credentials are built specifically for use with a particular moderator and do not require public verifiability or anchor concealment. This flexibility allows a wide range of cryptographic primitives—ranging from traditional blind signatures to anonymous credit tokens and rate-limited credentials—to instantiate MOLE credentials effectively.

A Feast of Cryptography and the Challenge of Parameter Selection

When designing complex cryptographic protocols, adaptability and efficiency are frequently at odds. While tailoring a construction for a specific purpose yields optimal performance, it drastically reduces the ability to adapt that same construction to alternative use cases. Adaptability, however, is an absolute prerequisite for a system like MOLE, which must support multiple credential types that will inevitably evolve over time to meet shifting internet traffic patterns and adversarial behaviors.

Most widely used post-quantum cryptographic schemes derive from larger mathematical families but are heavily tuned for singular applications. ML-DSA is a prime example; while its underlying mathematical framework is versatile, its specific parameters were carefully selected to build an efficient and secure standalone digital signature scheme. Those exact parameters would not necessarily maintain security if applied to entirely different cryptographic contexts.

This dynamic highlights the central challenge of designing post-quantum anonymous credentials: the necessity of decoupling parameter selection from protocol design. Classical credential systems enjoy this luxury naturally, allowing protocol designers and non-experts alike to build functional tools without worrying about the underlying mathematical parameters of an elliptic curve. Unfortunately, engineering efficient post-quantum equivalents has historically demanded an extraordinary level of specialized expertise. Fortunately, a recent breakthrough in cryptographic research promises to shift this paradigm entirely.

The post-quantum future of web privacy

VOLE-in-the-Head and the Rise of Multivariate Signatures

The conceptual roots of this recent development trace back to the introduction of the VOLE-in-the-Head zero-knowledge proof system. Zero-knowledge proofs enable a prover to demonstrate knowledge of a hidden witness satisfying a public statement without revealing any information about the witness itself. For instance, a proof can demonstrate knowledge of a secret encryption key corresponding to a known public output. These proofs can subsequently be transformed into digital signatures whose security relies directly on underlying cryptographic primitives, forming the basis for advanced signature candidates currently under evaluation by the National Institute of Standards and Technology.

Compared to many contemporary zero-knowledge architectures, VOLE-in-the-Head remains relatively straightforward to understand and implement securely. The underlying technique involves interactive processes that provide correlated random values acting as authentication tags, which a verifier uses to check the correctness of computations performed on a hidden witness. Provers locally prepare values that two parties would normally generate interactively, commit to those values, and utilize the Fiat-Shamir transformation to convert the interactive process into a non-interactive zero-knowledge proof.

While VOLE-in-the-Head is not succinct in the manner of other modern proof systems—meaning its proof sizes are asymptotically linear relative to the witness size and statement complexity—it is concretely efficient for statements expressed as equations over small binary fields. A prominent example is the KuMQuat signature scheme, where a signer proves an instance of the Multivariate Quadratic problem. This involves systems of random quadratic equations featuring multiple variables over small finite fields. Although the underlying multivariate mathematical problem is widely believed to remain secure against quantum computing attacks, proving knowledge of a solution via VOLE-in-the-Head is computationally inexpensive, yielding compact signatures.

Another established method for constructing multivariate signatures involves embedding specialized mathematical structures directly into the public map. The Unbalanced Oil and Vinegar signature scheme utilizes a trapdoor mechanism that allows authorized signers to easily solve multivariate quadratic equations by partitioning variables into distinct categories, while completely hiding this internal structure from the public key. While UOV offers remarkably small signature sizes, it traditionally suffers from exceptionally large public keys. Alternative multivariate schemes like MAYO attempt to balance this by significantly reducing public key sizes at the expense of slightly larger signatures.

Because the arithmetic complexity required to evaluate a multivariate signature is directly comparable to evaluating a general quadratic instance over small fields, these schemes integrate seamlessly with VOLE-in-the-Head frameworks. This synergy culminated in the publication of a groundbreaking blind signature scheme known as PoMFRIT, which achieved record-breaking post-quantum metrics for compact blind signatures and efficient issuance phases. By combining VOLE-in-the-Head with multivariate structures like MAYO or UOV, researchers have demonstrated that post-quantum anonymous credentials can achieve performance levels highly competitive with state-of-the-art lattice-based constructions.

Moving from Blind Signatures to Advanced Endorsements

Traditional blind signature designs typically require a client to place a message inside a cryptographic commitment—the mathematical equivalent of a sealed envelope—by combining a message hash with client-generated randomness. The issuer then signs this commitment directly. To construct the final blind signature, the client generates a zero-knowledge proof demonstrating knowledge of both the valid signature and the randomness used to open the commitment.

Unlinkability is naturally preserved because the client presents entirely different objects during each phase: a commitment is shown to the issuer during issuance, while a completely separate zero-knowledge proof is presented to a verifier during redemption. Even if the issuer and verifier collude, they possess no mechanism to link the two sessions.

For MOLE endorsements, the cryptographic requirements expand slightly to accommodate anchor-hiding properties. Instead of proving validity against a single public key, the protocol proves that a signature corresponds to one of multiple anchor keys within a trusted anchor set, treating the specific index of the anchoring key as part of the hidden witness. By carefully structuring the mathematical constraints, researchers can evaluate these proofs with minimal overhead, maintaining the core efficiency of the underlying hash and multivariate evaluations.

Furthermore, this modular framework easily extends to support advanced credential systems such as anonymous credit tokens and rate-limited credentials. In each scenario, the primary computational cost remains tied to the execution and evaluation of optimized hash functions, meaning any reduction in hashing overhead yields universal performance improvements across the entire credential ecosystem.

Optimizing for Efficiency and Confronting Security Assumptions

As researchers continue refining these post-quantum constructions, ongoing cryptanalysis of multivariate cryptography remains vital. While multivariate schemes such as UOV and MAYO have successfully weathered recent cryptographic attacks, parameter adjustments are frequently necessary to maintain robust security margins in the face of evolving analytical techniques.

To maximize real-world performance, engineers frequently employ targeted optimizations. One approach involves splitting proof generation into distinct stages to eliminate redundant hash evaluations inside the zero-knowledge circuit, allowing verifiers to compute hashes directly. While this substantially reduces overall proof sizes, it requires careful evaluation of underlying cryptographic assumptions to ensure that the removal of redundant verification steps does not inadvertently introduce new vulnerabilities.

Additional efficiency gains can be achieved by utilizing optimized hash functions and reducing internal state sizes. For instance, employing reduced-round variants of the Keccak permutation, such as TurboSHAKE, significantly trims the size of VOLE-in-the-Head proofs without sacrificing an acceptable security margin. Similarly, scaling down internal state dimensions reduces the number of active multiplications required within the proof circuit, shrinking the final cryptographic footprint.

Is MOLE Ready for Post-Quantum Deployment?

Current estimations for newly developed multivariate schemes—such as VOLE-IHAT for endorsements and VOLE-ACT for credentials—indicate that while post-quantum credentials remain notably larger than their classical predecessors, the overhead is entirely manageable. The operational runtimes for these protocols require only a few milliseconds per operation, proving that the technology is mathematically viable for production environments.

Although these post-quantum artifacts are larger than legacy standards, replacing traditional RSA blind signatures with schemes like PoMFRIT introduces operational overhead comparable to transitioning from ECDSA to ML-DSA in standard TLS environments. Because individual operations execute rapidly, the primary engineering challenge shifts away from fundamental cryptographic design and toward practical protocol engineering. For instance, rather than redeeming heavy anonymous credentials on every single HTTP request, web architects will need to design systems that efficiently amortize a single token redemption across multiple sequential requests.

With foundational research progressing rapidly and preliminary implementations showing strong performance, the transition to post-quantum anonymous credentials on the web is transitioning from theoretical cryptography into real-world application engineering. As ongoing cryptanalysis stabilizes parameters and researchers refine these constructions further, multivariate cryptography is poised to secure a foundational role in the future of privacy-preserving internet infrastructure.