Vitalik Buterin has urged cryptocurrency users to take AI-related threats to cryptographic security seriously, while warning them not to rush into wallet migrations that could create losses through operational mistakes.
The Ethereum co-founder’s comments came as researchers discussed whether advances in artificial intelligence could accelerate mathematical discoveries capable of weakening the systems that protect Bitcoin, Ethereum and other digital networks. According to CryptoSlate, Buterin said on Oct. 7 that users should reduce exposure where practical, but should not “scramble” to move funds into new wallets without a clear need.
The issue remains a warning about a possible future risk, not evidence that blockchain cryptography has already been broken. OpenAI recently published new mathematical results generated by an internal frontier model, including proofs formalized in Lean, but it did not report an attack on blockchain security. Neither Buterin nor Ethereum researcher Justin Drake cited evidence that ECDSA, the signature system used by Bitcoin and Ethereum, has been practically defeated.
Why AI has intensified the cryptography debate
Drake had called on the industry to start preparing for what he described as “bunker mode.” One measure would involve moving assets to fresh addresses whose public keys have never been exposed onchain. His concern is that AI could help researchers advance mathematical attacks on elliptic-curve cryptography, the family of techniques underpinning ECDSA.
Buterin broadened the discussion beyond existing signatures. He identified lattice-based cryptography, including ML-DSA, as another area where AI-assisted mathematical progress could reduce expected security margins over the next two years. ML-DSA was standardized by the National Institute of Standards and Technology in 2024 as a post-quantum digital-signature algorithm.
Lattice-based systems are important because they are among the leading candidates for protecting digital communications against sufficiently powerful quantum computers. Buterin’s warning is that the mathematical assumptions supporting these systems may also contain weaknesses that researchers have not yet found.
He pointed to the history of integer factorization as an example of how mathematical progress can alter security expectations. Techniques such as the general number field sieve reduced the effort needed to attack RSA compared with simpler approaches, prompting the use of much larger keys. In Buterin’s view, AI could compress what might otherwise be decades of mathematical development into a far shorter period, potentially improving attacks on elliptic-curve or lattice-based problems.
Ethereum favors designs with less mathematical structure
The concern is influencing the direction of Ethereum’s cryptographic planning. Buterin said the network’s “lean” roadmap has increasingly favored systems based primarily on hash functions over the past year.
That approach avoids lattice-based signatures such as ML-DSA and Falcon and reduces reliance on lattice-based commitments within zero-knowledge proofs. Hash-based signature systems, including WOTS and SPHINCS, are receiving more attention because they depend less on complex mathematical structures that could offer new opportunities for AI-assisted discoveries.
Hash functions are designed to expose comparatively little structure. That does not make them a universal replacement for public-key cryptography. Secure messaging, anonymous communications, websites, VPNs, Tor and encrypted web traffic use systems that require mathematical structures capable of creating a trapdoor: information that allows authorized users to decrypt or authenticate data.
Buterin said those systems could eventually require substantially larger security parameters if AI significantly improves cryptanalysis. He suggested that multiplying key sizes by as much as ten could become reasonable in such a scenario, although the source material does not describe this as a current requirement.
The potential implications therefore extend beyond digital-asset wallets. Any service relying on public-key encryption or digital signatures could face a trade-off between stronger security margins and the performance, storage and bandwidth costs of larger keys.
What crypto users can do—and why haste may be risky
For holders, the practical precautions described by Buterin are more limited than an immediate industry-wide migration. An address that has never signed a transaction can provide an additional layer of protection because its public key remains concealed behind the address hash. Once an account sends a transaction, its public key becomes visible.
Multisignature operators may also limit exposure by collecting signatures offchain and rotating signer keys after operations. This can reduce the period during which an exposed key might remain useful if ECDSA security were to deteriorate.
Buterin nevertheless cautioned against treating those measures as an emergency order to move funds. He said he had personally lost more money through failed migrations than through hacks. A rushed transfer can introduce configuration errors, lost access, incorrect addresses or other operational problems, even when the underlying cryptography remains intact.
The challenge for wallet providers, custodians and protocol developers is consequently two-sided: they must prepare for cryptographic risks that could develop faster than expected, while ensuring that security upgrades do not create more immediate failures.
Moneyiar market picture
Moneyiar data at 09:29 on Oct. 8, 2026, showed Bitcoin at $82,983.189054874, down 0.98% over 24 hours and 0.87% over seven days. Ethereum traded at $2,561.3047272397, falling 1.68% in 24 hours and 4.86% over seven days.
Shiba Inu was priced at $0.0000054083, down 1.76% over 24 hours and 5.9% over seven days. Uniswap stood at $7.7906146794, down 3.85% over 24 hours and 14.23% over seven days. These market movements are presented as Moneyiar data and are not identified in the source material as being caused by the cryptography debate.
