Buterin wrote that he doesn't recommend hastily transferring funds to new wallets, but the risks to cryptography posed by AI— which accelerates mathematical discoveries—should be taken seriously. Not only schemes vulnerable to quantum computers are at risk, but also those that might not withstand attacks enabled by AI. He identified lattice cryptography—including the post-quantum standard ML-DSA—and fully homomorphic encryption as new risk areas. And ECDSA, on which wallets rely, could fall even faster than expected.
The logic is this: over decades, mathematicians have found ways to break RSA much faster than initially thought, so keys had to be significantly lengthened. If AI can accomplish in two years what took 50 years of mathematical research, similar hidden vulnerabilities could emerge in elliptic curves and lattices as well. Hashes are more secure because they lack the mathematical structure that could be exploited. That’s why Ethereum has been moving for a year now toward signatures and ZK proofs based solely on hashes.
Practical advice: if possible, keep funds on addresses from which no transactions have yet been sent, and collect multi-signature confirmations off-chain to avoid revealing signatures. And be cautious when migrating: “Personally, I’ve lost more money on unsuccessful migrations than on all hacks combined.”