In April 2008, an auction house in New York offered a run of Domaine Ponsot Clos Saint-Denis from the 1940s. The bottles looked right. The labels were right. The capsules were right. Minutes before the lots came up, Laurent Ponsot had them pulled. His family's domaine had not made that wine until 1982.
That was the beginning of the end for Rudy Kurniawan, who was later sentenced to ten years for the most famous wine fraud of the century. What stayed with me, from years in auction rooms and cellars since, is not how good the forgeries were. It's how they were caught. Nobody spotted a bad label. Someone knew a history that couldn't be true.
A label is a kind of signature. It says: this came from where I say it came from. And every signature is only as good as the cost of faking it. For a long time a Ponsot label was expensive enough to fake, and the market thin enough, that nobody tried. Then someone did, at scale, and the trade learned that the label had never been the protection. The protection was the chain: who held the bottle, where it slept, who could vouch for every step.
I've been thinking about that this month, because the people who secure the internet have started having the same conversation.
The warning
Almost everything you do online rests on a signature. When your bank confirms it's really your bank, when a payment goes through, when a crypto wallet moves funds, a piece of mathematics is doing the job a label does: it's cheap to check and, as far as anyone knows, practically impossible to forge.
On October 7, Justin Drake, a researcher at the Ethereum Foundation, wrote that this assumption might not hold much longer. The trigger was a release from OpenAI the day before: hundreds of new mathematical manuscripts produced by one of its internal models. Drake's worry is that a sufficiently capable AI could find a shortcut through the mathematics behind the signatures that protect bitcoin and ether, sooner than the quantum computers everyone has been watching for. In the worst case, he wrote, months rather than years.
His advice has a name, "bunker mode", and it's simple. On these networks, the public half of your key is only revealed the first time you sign something with it. So move your funds to an address that has never signed anything. Don't show the forger your label.
To be clear about what this is: no one has broken anything. Drake called it his personal advice, not an alarm. Vitalik Buterin agreed the risk is real and told people not to scramble. Others called it fear-mongering. All of that is fair. But serious people are now saying it out loud, and that's new.
Why it sounds like science fiction
The same week, Andrej Karpathy, one of the people who built modern AI, described why conversations like this go nowhere. Most of the world, he estimated, has barely touched these systems. About one in five people have tried a free chatbot and seen something clumsy. One in five hundred use the best models for serious technical work every day, and watch weeks of work get done from a single instruction. And a few thousand people inside the labs see something further along again, which he described in words that should make anyone in security sit up: agents minting new software exploits at machine speed.
If your only experience of AI is a chatbot fumbling a simple question, a warning about it breaking the mathematics of money sounds absurd. If you sit near the top of that funnel, it sounds like Tuesday.
The wine trade has been here. For years, most collectors had never knowingly met a fake. The few people who opened enough old Burgundy to know what the corks should look like had been uneasy long before 2008. The warning always comes from the people closest to the work, and it always sounds like paranoia to everyone else.
Where I come down
Here's my view, and it's the part I find more interesting than the fear.
This is not only crypto's problem. The same family of mathematics protects the bank, the website, the payment terminal at the end of the bar. If it breaks, it breaks everywhere. Crypto just has the least room for error, because a stolen coin can't be called back the way a bank reverses a fraudulent transfer.
But that's also why I think crypto may end up leading the response rather than suffering it. It's the one system that has spent fifteen years under constant attack, in public, with real money as the bounty. It's where much of the next generation of cryptography has already been built and tested. And it's a system that can change its own rules in the open, where everyone can watch the fix go in. If the mathematics of trust needs rebuilding, it's not obvious to me that a closed institution patching itself in private is the better place to start.
There's a phrase going around for this: secure the money, secure the future. I'd put it more modestly. The industry that learned to trust nothing may be the one best placed to rebuild trust for everyone else.
What the cellar already knows
The lesson from Kurniawan wasn't "make better labels." It was that the label was never enough. Provenance is a chain, not a mark. You trust a bottle because you can account for every hand it passed through, and because the people at the end of that chain knew enough history to say: that can't be true.
Bunker mode is the same lesson, in mathematics. Don't rely on the one mark a forger might learn to copy. Keep the chain unbroken. Know where everything you value has been.
And if you run a business, it's worth asking a question most of us never have: every system you trust with your money is signed by mathematics you've never examined. Somebody should know what the plan is if that changes. It might as well be you.