Vitalik Buterin flags AI as a new threat to crypto's math foundations
Regulation & Gov ·
Ethereum's co-creator says machine-driven mathematical progress could erode both lattice cryptography and elliptic-curve signatures sooner than engineers have planned for.
In a post on X, Buterin argued that the next two years of AI-assisted mathematics could do to lattice-based systems what decades of human research did to factoring: compress a once-secure parameter into something crackable, the way the shift from naive factoring to the number field sieve shrank RSA's effective safety margin and forced key sizes up toward 400 bytes, according to his statement. He named ML-DSA, fully homomorphic encryption schemes, and lattice-based commitments used in zero-knowledge proofs as the areas carrying fresh exposure, on top of the long-known quantum risk to ECDSA's elliptic curves.
That reasoning, he said, is why Ethereum's lean roadmap has spent the past year steering toward hash-based cryptography — WOTS and SPHINCS+ signatures — instead of lattice or Falcon schemes, since hash functions are not believed to carry the kind of exploitable algebraic structure that a sufficiently advanced AI could target. The harder problem, he noted, is public-key encryption itself, which mathematically cannot be built from hashes alone and requires some structured trapdoor — group theory, isogenies, lattices, or code-based constructions — any of which he said should be assumed vulnerable to at least partial AI-driven attack, suggesting that key sizes may need to grow by a factor of ten to stay safe.
Buterin was explicit that the warning is not a call to action: he does not recommend rushing funds into new wallets, framing the "fresh address" habit as a precaution rather than an emergency measure, a point echoed in reporting from CryptoPotato and The Block, both of which described his position as urging caution over panic.
The cluster of coverage, including accounts from WuBlockchain and Leviathan News, consistently frames the remarks as a bearish signal on lattice cryptography's medium-term durability rather than an announcement of any imminent break. Buterin himself noted the probability that hash-based schemes could also fail is tied to whether P equals NP, which he considers very unlikely.
What remains unresolved is the timeline: Buterin offers a plausibility argument rather than a specific forecast, and no concrete attack or proof-of-concept against lattice or ECDSA systems has been reported. Whether Ethereum's shift toward hash-only signatures expands further, and whether other protocols relying on lattice-based or elliptic-curve primitives begin similar migrations, is the open question going forward.