The Decoder reported on 8 October that two leading Ethereum researchers see a worst case where AI-assisted maths breaks wallet signatures within months. Justin Drake wrote on X that the industry should prepare a "bunker mode": keep funds at addresses that have never signed anything. A signature reveals the public key. In theory, an attacker could work backward from it to the private key. Vitalik Buterin, Ethereum's co-founder, replied on X that he agreed but urged patience.
Buterin said migration mistakes have cost him more than hacks have. Long term, he wants designs built mainly on hash functions, which are one-way scrambles of data. He argues AI advances could weaken even lattice-based schemes, which many see as quantum-safe. The Decoder says nobody has broken the current ECDSA scheme in practice. It does not name the maths results behind the warning, explain the "months" estimate, or give a date.
Interpretation, not from the source: the choice is between a theoretical threat and a migration risk that is certain. Moving funds off an address that has already signed needs a new signature, so the move itself may be a weak point. Custody and wallet teams could ask three things. How many of our addresses have already shown their public keys? Who controls a migration? What is the rollback plan?