Justin Drake’s call for “bunker mode” and mass migration, alongside talk of cryptographic breaks within months, is hyperbolic. There is no evidence of a new attack on elliptic curves or any quantitative basis for a short-term timetable for such attacks emerging. Prominent researchers need to consider how non-experts will receive their language. Speculating that within months individuals could break cryptographic primitives that the internet and digital economy rely on carries consequences. It is a huge leap from AI making progress in other fields of mathematics to predicting the imminent failure of primitives that have faced decades of cryptanalysis. The idea that hash functions are safer from AI than elliptic curves is speculation. Hash functions contain structure, and we model them as random oracles without justifying proofs. There is no evidence that elliptic curve cryptography is more vulnerable to AI than symmetric cryptography. If historical human-led analysis is any predictor of the future, the opposite is more likely. The worst possible reaction to AI developments is to rush into new cryptographic primitives, away from multi-decade old hardened primitives, because somebody is claiming they are safe. I worry that this is an inevitable consequence of the current discourse. Because of the last 48 hours users are worrying about speculative non-existent black hats, who apparently have the capability of signing OS upgrades and forging bank credentials, from draining their wallets. To protect against speculative threats Ethereum is pulling talent and attention into base-layer primitives and technological iteration, instead of increasing demand for block space by building compelling applications users want and love. I don’t want to dismiss the considerable advances in mathematics that AI is driving. It is important to understand the nature of these improvements. If its success in other mathematical fields applies to elliptic curves, my expectation is that AI will grind its way to finding algorithms that improve on the current state of the art. That could mean shaving off a few bits of security. It would be prudent to discuss whether we need to increase key sizes and by how much. I can’t guarantee that AI won’t find total breaks, and neither can anyone else. My speculation as a domain expert does not make my speculation authoritative. We should also be using AI to investigate whether searching the parameter space can yield new, more secure curves. We can investigate whether some of these “structure = bad” hunches have any basis in reality. For example, we could use AI to see if the extra structure in Koblitz curves enables attacks beyond the small speedups we know about. This would give us actual evidence to make informed decisions. If non-experts hear that the cryptography underpinning the digital economy could fail within months, the rational response is to withdraw all digital money (on-chain and off) and hoard gold under the bed. I think the past 48 hours has unnecessarily undermined confidence in the entire concept of blockchains and digital money.