Vitalik Buterin warns AI could weaken the math behind crypto security

Most crypto security debates fixate on quantum computers that may arrive someday. Vitalik Buterin just pointed at a threat that might show up sooner: AI that is very good at math.

On October 7, 2026, the Ethereum co-founder warned that AI-accelerated mathematical research has a “good chance” of significantly weakening lattice-based cryptography within two years. That family includes ML-DSA and fully homomorphic encryption (FHE). He also flagged added risk for ECDSA, the widely used elliptic curve signature algorithm.

What Buterin actually said

Lattice cryptography is the main target of his warning. It relies on hard geometric problems involving grids of points in many dimensions. ML-DSA is a lattice-based signature scheme. FHE is a technique that lets computers process encrypted data without ever decrypting it, which is useful for privacy applications.

Buterin paired the warning with a set of practical recommendations:

• Bigger lattice parameters. Where lattice schemes are still used, he suggested conservative settings, including a 10-fold increase in key sizes where applicable.
• No encrypted data stored directly on-chain. Data posted to a blockchain stays there permanently. If the encryption protecting it weakens later, the privacy is gone for good.
• Key rotation for multisigs. He said multisig signers should switch to new keys after each operation.
• Offchain signature gathering. Multisigs should collect signatures offchain where possible to cut down on exposure.
• Caution on mass fund migrations. He discouraged large-scale moves of funds, which carry their own risks.

The multisig advice connects to a broader point he made about public keys. Exposing a public key on-chain gives future attackers something to work with. As AI and cryptanalysis techniques advance, that exposure becomes more of a liability.

Ethereum’s roadmap shifts toward hashes

Buterin described a move away from depending on lattice cryptography and toward pure hash-based signature schemes, naming WOTS and SPHINCS+. Hash-based signatures rest on a simpler foundation, with security leaning on hash functions. Fewer exotic math assumptions means fewer places for a clever new attack to land.

WOTS, short for Winternitz one-time signature, is designed so that each key signs only once. That design philosophy rhymes with his multisig guidance about rotating keys after each operation.

In September 2026, Buterin had projected that AI-assisted formal verification could benefit cybersecurity. Formal verification uses mathematical proofs to check that code behaves exactly as intended. The same technology he saw as a defensive upgrade now looks like it could cut both ways: AI that helps prove code is correct may also help prove that certain encryption is weaker than advertised.

Background: a crowded threat calendar

Lattice schemes have been widely discussed as candidates for resisting quantum attacks. If AI-driven math research erodes them first, protocols could find themselves migrating away from the very tools meant to be the safe harbor.

Buterin keeps the majority of his net worth in crypto assets, which puts his own holdings on the line if signature schemes falter. His caution about large-scale migrations reads less like abstract theory and more like a risk he is managing himself.

What this means for users, builders and investors

No immediate price reactions or protocol exploits were reported after the warning. The focus has stayed on long-term cryptographic strategy rather than any acute incident.

For builders, teams relying on lattice-based tools, including FHE for privacy features, may want to revisit parameter choices. Projects tempted to store encrypted data on-chain now have a prominent voice arguing against it.

For multisig operators, such as treasuries, DAOs and custody setups, the advice is operational. Rotate keys after each use. Gather signatures offchain. Minimize how often public keys hit the chain.

Buterin specifically warned that large-scale migrations carry their own risks. A rushed transfer can create the very exposure people are trying to avoid.

Ethereum’s shift toward hash-based signatures like WOTS and SPHINCS+ is a long project, and Buterin’s window for lattice weakening is two years. Whether the roadmap moves faster than the math research it is racing against is the variable worth watching.