The Mountain in the Fog
An AI just produced mathematics that the mathematicians who read it can't fully follow. One of them said it felt like being teleported to a mountaintop in fog — the view is real, but she has no idea how she got there. That feeling is the first serious threat to the one principle this whole publication is built on.
This week, by the account going around, an AI lab released hundreds of new mathematical results at once — enough that a serious theoretical computer scientist called it, in the same breath, "one of the biggest days in mathematical history" and a "mathpocalypse." The reactions from working mathematicians are the part worth sitting with. One described reading the output as being teleported to the top of a mountain, surrounded by fog. She could see the view. She just couldn't see the path — couldn't retrace how the machine got there, couldn't make the steps her own.
An institute of advanced study put the same worry more flatly: it is now the case that an AI can output a mathematical argument that the human who prompted it cannot understand, cannot verify, and cannot take responsibility for.
Read that twice, because it's the sentence that should stop you. Not because math is scary. Because verification is the thing this entire newsletter has been about — and something just produced a truth that no human could check.
The principle that just got threatened
The deepest idea in sound money, the one under everything, is three words: don't trust, verify. You don't take the central bank's word for the money supply — you check it. You don't trust the custodian — you hold the keys. The whole project was to replace faith in an authority with proof you can confirm yourself. Verification was the exit from trust.
Now look at what the mountain in the fog does to that. If an answer arrives that is correct but that no human can follow, verification quietly dies — and trust walks straight back in through the door we thought we'd closed. You're handed a result. You cannot check it. So you either reject it or you trust the oracle that produced it. There is no third option, because the third option was understanding, and understanding is exactly what the fog took.
This is the cow and the machine, arriving at the one place it hurts most. In an earlier essay I argued that "verify, don't trust" only works as long as "yourself" is enough that against a superhuman mind, checking-it-yourself quietly becomes faith in a lab coat. Mathematics was supposed to be the last redoubt, the one domain where proof meant anyone could follow the steps and be certain. A machine that produces proofs humans can't retrace takes even that. The cow can now read the theorem and remain, in the only sense that matters, a cow.
The concrete version: the locks on everything
The abstract threat has a very concrete edge, and it's why cryptographers are the ones sounding alarms this week. One called it time for "bunker mode." Another said, simply, "I think we might lose public-key cryptography."
Here's what that means, in plain terms, because most people don't know how much rides on it. Almost everything you do online — logging in, banking, shopping, messaging — is secured by public-key cryptography. It works because certain math problems are easy to do one way and effectively impossible to reverse: easy to lock, impossibly hard to pick. Your security is the gap between those two. The entire digital world is built on the assumption that nobody, and nothing, can close that gap.
A superhuman mathematician is precisely a machine for closing gaps like that. If an AI discovers the math that reverses the one-way problem, the locks don't get harder to pick. They stop being locks.
And here's the Bitcoin-specific nuance, because it's genuinely clarifying. A Bitcoin address is not your public key — it's a hash of it, a different and tougher kind of lock. Your public key stays hidden behind that hash until the moment you spend, which is when it's finally revealed. That's why the researcher's advice was to move coins to fresh addresses that have never exposed a public key: coins sitting behind an unused hash are shielded by a lock that this kind of attack doesn't directly pick, while coins whose public key has already been shown to the world are the exposed ones. It's a real distinction, and it tells you the signature layer is where the danger is, not the hashing that secures mining.
The inversion nobody leads with
Now the part the "crypto investors, panic" framing gets backwards.
If public-key cryptography actually falls, Bitcoin is the least of what breaks. The same math secures your bank, your email, your government's secrets, the login to every account you own, the padlock on every website. "Move your crypto to a bunker" dramatically undersells it. The honest headline is: everyone who uses the internet should be wondering, and the newsletter's own last line admits as much. Bitcoin is a rounding error next to the whole digital world unlocking at once. And then the genuinely counterintuitive turn, the one that ties back to this publication's spine. Of all the systems exposed, the decentralized one may be among the best placed to survive — because it has something the sprawling legacy internet doesn't: a clear way to change its own rules. Cryptographers have been building post-quantum signature schemes — new locks designed to resist exactly this kind of attack — for years. The question was never whether replacements exist. It's whether a system can coordinate the migration to them. And as the essay on Bitcoin's games argued, Bitcoin has a working mechanism for changing its rules by consensus when the incentives align — and nothing aligns incentives like "upgrade or every coin dies." A thing with no central gate also has no single point that must approve the fix. The castle, by contrast — ten thousand banks, agencies, and legacy systems, each with its own gate — has to coordinate a migration no one is in charge of. The decentralized thing might adapt faster than the centralized ones, not slower. The drawbridge cuts both ways: a castle is also a thing that can't move.
The reassurance, and the part it doesn't reach
Breathe, because the panic is ahead of the facts, and the facts are worth stating. The lab's own results, by the account, solved only about 5% of the problems it attempted a failing grade anywhere. And the proofs it did produce were described by one mathematician as reading "like something written by someone on psychedelics" unclear, not making sense, correct-looking without being comprehensible.
Hold onto why that's reassuring, because it's the whole distinction. The machine is producing answers without understanding — the result without the path, the view without the climb. And that is the cargo cult from an earlier essay, running in reverse: all the forms of a proof, the symbols and the structure, wrapped around something no human mind can yet inhabit. As long as understanding is the thing that counts — as long as a truth you can't follow is treated as suspect rather than authoritative — humans are still, in the way that matters, better at mathematics than machines. For now.
But notice what that reassurance does not reach. Patch the cryptography — swap every lock for a post-quantum one, migrate the coins, save the banks — and the fog is still on the mountain. The deeper thing hasn't been fixed, because it isn't a bug to fix. A machine that produces truths humans cannot verify is a permanent new condition, and it forces a question the species has never had to answer: will we accept answers we cannot check? Every instinct trained by "verify, don't trust" says no. Every incentive speed, power, the fusion reactor and the cancer cure the same machine might hand us — says yes. And the honest truth is that we will probably say yes, piece by piece, because the view from the mountain is too good to refuse just because we can't see the path. That is the actual mathpocalypse. Not that the locks might break — locks can be replaced. It's that verification itself, the exit from trust that this whole publication was built to celebrate, may turn out to have been a brief human window that is quietly closing: a few centuries where the proofs were simple enough that anyone could follow them, ending in a fog.
The one line to keep
An AI produced mathematics its own readers couldn't retrace, and the tremor ran straight to the foundation — because "verify, don't trust" only stands while verifying is something a human can still do.
The concrete danger is real and, oddly, the most fixable part: public-key cryptography secures everything, not just Bitcoin, and if it falls the whole internet falls with it — but new locks exist, and the decentralized systems with a way to change their own rules may migrate to them faster than the castle ever could. Hold that line where you can. Move to fresh keys, back the upgrade, don't panic about the one asset while ignoring that your bank runs on the same math.
But keep your eye on the fog, because it's the real frontier. The machine that does mathematics we cannot check is the first thing that could march us back from verification to trust — not by breaking a lock, but by handing us gifts we can use and never understand. The question for anyone who uses money, or the internet, or their own mind, is no longer only can I verify this. It's becoming: what will I do when I no longer can?
The view from the mountain is extraordinary. Just remember you didn't climb it — and that forgetting the difference is how you get lost up there.
An AI just produced mathematics that the mathematicians who read it can't fully follow. One of them said it felt like being teleported to a mountaintop in fog — the view is real, but she has no idea how she got there. That feeling is the first serious threat to the one principle this whole publication is built on.
This week, by the account going around, an AI lab released hundreds of new mathematical results at once — enough that a serious theoretical computer scientist called it, in the same breath, "one of the biggest days in mathematical history" and a "mathpocalypse." The reactions from working mathematicians are the part worth sitting with. One described reading the output as being teleported to the top of a mountain, surrounded by fog. She could see the view. She just couldn't see the path — couldn't retrace how the machine got there, couldn't make the steps her own.
An institute of advanced study put the same worry more flatly: it is now the case that an AI can output a mathematical argument that the human who prompted it cannot understand, cannot verify, and cannot take responsibility for.
Read that twice, because it's the sentence that should stop you. Not because math is scary. Because verification is the thing this entire newsletter has been about — and something just produced a truth that no human could check.
The principle that just got threatened
The deepest idea in sound money, the one under everything, is three words: don't trust, verify. You don't take the central bank's word for the money supply — you check it. You don't trust the custodian — you hold the keys. The whole project was to replace faith in an authority with proof you can confirm yourself. Verification was the exit from trust.
Now look at what the mountain in the fog does to that. If an answer arrives that is correct but that no human can follow, verification quietly dies — and trust walks straight back in through the door we thought we'd closed. You're handed a result. You cannot check it. So you either reject it or you trust the oracle that produced it. There is no third option, because the third option was understanding, and understanding is exactly what the fog took.
This is the cow and the machine, arriving at the one place it hurts most. In an earlier essay I argued that "verify, don't trust" only works as long as "yourself" is enough that against a superhuman mind, checking-it-yourself quietly becomes faith in a lab coat. Mathematics was supposed to be the last redoubt, the one domain where proof meant anyone could follow the steps and be certain. A machine that produces proofs humans can't retrace takes even that. The cow can now read the theorem and remain, in the only sense that matters, a cow.
The concrete version: the locks on everything
The abstract threat has a very concrete edge, and it's why cryptographers are the ones sounding alarms this week. One called it time for "bunker mode." Another said, simply, "I think we might lose public-key cryptography."
Here's what that means, in plain terms, because most people don't know how much rides on it. Almost everything you do online — logging in, banking, shopping, messaging — is secured by public-key cryptography. It works because certain math problems are easy to do one way and effectively impossible to reverse: easy to lock, impossibly hard to pick. Your security is the gap between those two. The entire digital world is built on the assumption that nobody, and nothing, can close that gap.
A superhuman mathematician is precisely a machine for closing gaps like that. If an AI discovers the math that reverses the one-way problem, the locks don't get harder to pick. They stop being locks.
And here's the Bitcoin-specific nuance, because it's genuinely clarifying. A Bitcoin address is not your public key — it's a hash of it, a different and tougher kind of lock. Your public key stays hidden behind that hash until the moment you spend, which is when it's finally revealed. That's why the researcher's advice was to move coins to fresh addresses that have never exposed a public key: coins sitting behind an unused hash are shielded by a lock that this kind of attack doesn't directly pick, while coins whose public key has already been shown to the world are the exposed ones. It's a real distinction, and it tells you the signature layer is where the danger is, not the hashing that secures mining.
The inversion nobody leads with
Now the part the "crypto investors, panic" framing gets backwards.
If public-key cryptography actually falls, Bitcoin is the least of what breaks. The same math secures your bank, your email, your government's secrets, the login to every account you own, the padlock on every website. "Move your crypto to a bunker" dramatically undersells it. The honest headline is: everyone who uses the internet should be wondering, and the newsletter's own last line admits as much. Bitcoin is a rounding error next to the whole digital world unlocking at once. And then the genuinely counterintuitive turn, the one that ties back to this publication's spine. Of all the systems exposed, the decentralized one may be among the best placed to survive — because it has something the sprawling legacy internet doesn't: a clear way to change its own rules. Cryptographers have been building post-quantum signature schemes — new locks designed to resist exactly this kind of attack — for years. The question was never whether replacements exist. It's whether a system can coordinate the migration to them. And as the essay on Bitcoin's games argued, Bitcoin has a working mechanism for changing its rules by consensus when the incentives align — and nothing aligns incentives like "upgrade or every coin dies." A thing with no central gate also has no single point that must approve the fix. The castle, by contrast — ten thousand banks, agencies, and legacy systems, each with its own gate — has to coordinate a migration no one is in charge of. The decentralized thing might adapt faster than the centralized ones, not slower. The drawbridge cuts both ways: a castle is also a thing that can't move.
The reassurance, and the part it doesn't reach
Breathe, because the panic is ahead of the facts, and the facts are worth stating. The lab's own results, by the account, solved only about 5% of the problems it attempted a failing grade anywhere. And the proofs it did produce were described by one mathematician as reading "like something written by someone on psychedelics" unclear, not making sense, correct-looking without being comprehensible.
Hold onto why that's reassuring, because it's the whole distinction. The machine is producing answers without understanding — the result without the path, the view without the climb. And that is the cargo cult from an earlier essay, running in reverse: all the forms of a proof, the symbols and the structure, wrapped around something no human mind can yet inhabit. As long as understanding is the thing that counts — as long as a truth you can't follow is treated as suspect rather than authoritative — humans are still, in the way that matters, better at mathematics than machines. For now.
But notice what that reassurance does not reach. Patch the cryptography — swap every lock for a post-quantum one, migrate the coins, save the banks — and the fog is still on the mountain. The deeper thing hasn't been fixed, because it isn't a bug to fix. A machine that produces truths humans cannot verify is a permanent new condition, and it forces a question the species has never had to answer: will we accept answers we cannot check? Every instinct trained by "verify, don't trust" says no. Every incentive speed, power, the fusion reactor and the cancer cure the same machine might hand us — says yes. And the honest truth is that we will probably say yes, piece by piece, because the view from the mountain is too good to refuse just because we can't see the path. That is the actual mathpocalypse. Not that the locks might break — locks can be replaced. It's that verification itself, the exit from trust that this whole publication was built to celebrate, may turn out to have been a brief human window that is quietly closing: a few centuries where the proofs were simple enough that anyone could follow them, ending in a fog.
The one line to keep
An AI produced mathematics its own readers couldn't retrace, and the tremor ran straight to the foundation — because "verify, don't trust" only stands while verifying is something a human can still do.
The concrete danger is real and, oddly, the most fixable part: public-key cryptography secures everything, not just Bitcoin, and if it falls the whole internet falls with it — but new locks exist, and the decentralized systems with a way to change their own rules may migrate to them faster than the castle ever could. Hold that line where you can. Move to fresh keys, back the upgrade, don't panic about the one asset while ignoring that your bank runs on the same math.
But keep your eye on the fog, because it's the real frontier. The machine that does mathematics we cannot check is the first thing that could march us back from verification to trust — not by breaking a lock, but by handing us gifts we can use and never understand. The question for anyone who uses money, or the internet, or their own mind, is no longer only can I verify this. It's becoming: what will I do when I no longer can?
The view from the mountain is extraordinary. Just remember you didn't climb it — and that forgetting the difference is how you get lost up there.
1
