Ethereum has spent years adding increasingly sophisticated cryptography to improve scalability, privacy and security. Now Vitalik Buterin is arguing that one way to protect the network from future mathematical breakthroughs may be to move in the opposite direction. His concern is that artificial intelligence could accelerate mathematical research enough to uncover new attacks against cryptographic structures that currently appear secure. The counterintuitive response is not to abandon cryptography, but to reduce the number of complicated mathematical assumptions Ethereum needs to trust.

The warning follows rapid advances in AI-assisted mathematics. Buterin argued that lattice-based cryptography, including systems designed specifically for the post-quantum era, could suffer significant reductions in concrete security if AI discovers mathematical shortcuts faster than humans expect. His concern extends the threat model beyond quantum computers: a cryptographic problem does not need to become easy, only substantially easier than designers assumed when choosing its security parameters. That possibility has already reinforced Ethereum’s shift toward post-quantum security and AI-aware verification, where cryptographic resilience increasingly influences long-term protocol design.

The Problem Is Mathematical Structure

Modern cryptography often depends on mathematical problems that are easy to construct but extraordinarily difficult to reverse. Elliptic-curve cryptography, for example, relies on the difficulty of solving discrete logarithm problems, while lattice schemes depend on hard problems involving high-dimensional geometric structures. These assumptions are backed by decades of research, but they remain assumptions about what attackers cannot efficiently calculate. Buterin’s argument is that every additional mathematical structure gives future researchers, including AI systems, something potentially exploitable to study.

History makes that concern difficult to dismiss entirely. Improvements in integer-factorization algorithms dramatically changed the key sizes required to make RSA secure, even though factoring itself was never rendered trivial. Buterin’s analogy is that AI could potentially deliver similar improvements against elliptic curves or lattices, shrinking security margins without producing a spectacular overnight “break.” If AI compresses decades of mathematical experimentation into a few years, cryptographic systems may need to assume that unexpected algorithmic advances will arrive faster than previous security models anticipated.

That is particularly uncomfortable because lattice cryptography has become one of the principal defenses against quantum computing. NIST standardized the lattice-based ML-DSA alongside other post-quantum algorithms in 2024. Yet it also standardized SLH-DSA, a signature system based on hash functions rather than lattice problems. Post-quantum does not mean mathematically invulnerable; it means resistant to the attacks currently expected from quantum computers under the assumptions used today. NIST’s post-quantum standards include both lattice-based and hash-based approaches.

Ethereum’s response increasingly favors the second category where practical. Its Lean Ethereum research uses hash-based signatures such as WOTS and SPHINCS-derived constructions instead of depending on lattice signatures for validator authentication. The network’s broader post-quantum roadmap already includes hash-based replacements for vulnerable BLS validator signatures. The security philosophy is essentially to use cryptographic primitives with as little exploitable algebraic structure as possible when the application allows it.

Hashes Offer a Different Kind of Defensive Simplicity

A cryptographic hash function takes arbitrary data and produces a fixed-size output designed to reveal almost nothing about the original input. Its security still depends on assumptions, and no responsible cryptographer would claim that hashes are impossible to break. But they deliberately offer far less mathematical structure than systems based on elliptic curves or lattices.

That difference is central to Buterin’s reasoning. A highly structured mathematical object can contain properties that have not yet been discovered. AI could potentially search those relationships more effectively than human researchers and find shortcuts that reduce the work required for an attack. A well-designed hash function aims to give mathematics fewer handles to grab onto, making unexpected structural shortcuts less plausible even if they cannot be ruled out entirely.

Ethereum has already made engineering decisions consistent with that philosophy. Layer-1 research recently moved away from the specialized Poseidon hash toward established hash families such as SHA and BLAKE. The motivation is not simply that older primitives are fashionable or easier to understand. Widely deployed hashes have undergone years of cryptanalysis, while specialized cryptographic constructions introduce new assumptions that must themselves survive scrutiny.

There is an important cost, however. Hash-based signatures can be substantially larger than the BLS signatures Ethereum currently uses for consensus, and they lack the same convenient native aggregation properties. Ethereum therefore needs additional machinery such as leanVM to compress large numbers of signatures efficiently. Reducing cryptographic assumptions does not necessarily reduce engineering complexity; sometimes the protocol must become more sophisticated precisely so its underlying security assumptions can become simpler.

This tension also explains why Ethereum’s future architecture increasingly emphasizes cryptographic agility. Features associated with the planned post-Hegotá transition toward proof-based and post-quantum infrastructure would make it easier for accounts and protocol components to change signature schemes as security assumptions evolve. A blockchain expected to operate for decades cannot reasonably assume that the cryptography considered strongest in 2026 will remain optimal indefinitely.

Ethereum Cannot Become Hash-Only Everywhere

The strongest limitation to the thesis is that hashes cannot replace every kind of cryptography Ethereum and the broader internet require. Buterin specifically identified public-key encryption as the harder problem. Digital signatures and many proof systems can move toward hash-based constructions, but encryption between parties requires mechanisms that provide additional mathematical structure.

That means lattice systems, code-based cryptography or other structured approaches may remain necessary for applications such as secure communication, privacy protocols and certain encryption systems. Ethereum can minimize structured cryptography, but it cannot eliminate the need for mathematical structures simply because those structures create theoretical attack surfaces. Buterin’s suggested response for areas where they remain unavoidable is more conservative parameter selection, potentially accepting much larger keys and greater computational costs to preserve safety margins.

Privacy creates a particularly difficult trade-off. Encrypting information and publishing that ciphertext permanently onchain gives future attackers unlimited time to discover a weakness and decrypt historical data. Buterin has therefore argued that privacy systems should avoid placing encrypted notes onchain where possible, preferring offchain transmission mechanisms when the use case permits. This is an architectural security decision rather than simply a choice of stronger cipher. Sometimes reducing cryptographic risk means reducing how much valuable encrypted material the protocol exposes in the first place.

The same principle applies to wallets. Buterin has supported cautious preparation for a world in which exposed public keys become more dangerous, while warning users against rushed migrations that can themselves cause losses. The recent debate over AI-driven cryptographic risk and “bunker mode” is therefore less about evidence that Ethereum’s current keys are being cracked today and more about preparing migration paths before a real break forces everyone to act simultaneously.

Ethereum’s emerging strategy can consequently be understood as cryptographic minimalism rather than cryptographic retreat. The network still needs signatures, proofs, commitments and privacy technologies, but it can become more selective about the assumptions embedded in them. The safer long-term protocol may not be the one with the most advanced cryptography, but the one that depends on the smallest number of assumptions that future mathematics has an opportunity to overturn.

That distinction becomes more important if AI genuinely accelerates mathematical discovery. Ethereum cannot know which cryptographic primitive will fail first, nor whether the breakthrough will come from quantum hardware, artificial intelligence or ordinary human research. It can, however, design the protocol so fewer components depend on fragile assumptions and replacements can be introduced without rebuilding the network from scratch. Using “less cryptography” is therefore slightly misleading: Ethereum’s real objective is to use less cryptographic structure wherever simpler foundations can provide the same security function.