Paramedic 🔺

The debate on the threat AI poses for crypto is mostly just wankery. We will see AI exploit systems-level bugs far before ECDSA falls. We will see AI discover bugs in systems code, libraries and wallets that enable attackers to steal billions of dollars long before we have to worry about cryptosystems failing wholesale due to new math. Sure, there's definitely a non-zero possibility that ECDSA might fail wholesale, and the ramifications would be systemic -- all balances on such a chain would be in question. That's why Hoseeb's backup crypto proposal is sensible. Namely, build an alternative cryptosystem onto which one could shift should ECDSA doomsday arrive one fine morning. That's an approach that we will be exploring for Avalanche. But our focus regarding AI's impact on blockchains ought to center on vulnerabilities in existing software first and foremost. If you think that, say, Ripple's (or insert any other high value low use chain here) under scrutinized code contains no 0-day bugs, difficult for humans to spot but easy for AI to identify, then you're deluded. My take on this issue, as with all other issues, is pragmatic and oriented towards what matters most today. That said, in the tail risk category, hash-based crypto is really interesting, and we have an active group exploring hash-based alternatives to ECDSA.

Avalanche supports both the UTXO and the EVM account model. The account model has turned out to be incredibly useful overall, but isn't designed cleanly enough to limit state growth. The UTXO model has some security and (possibly) tax efficiency advantages, but is cumbersome and error prone.