Doug Colkitt - Will AI Crack Crypto, Bunker Mode, and the Y2K Comparison
- Doug Colkitt sees AI’s mathematical leap as a regime change in uncertainty, not evidence that crypto will break within months. Models progressed from SAT and Olympiad-style questions to Erdős problems, a claimed Navier–Stokes result, and OpenAI’s release of 400 open problems across multiple fields. The concern is that “all bets are off the table” when nobody knows how quickly long-held hardness assumptions might fall.
- Justin Drake’s “bunker mode” addresses a real cryptographic asymmetry, but Doug considers the prescription extreme today. Elliptic-curve and lattice systems may contain exploitable structure, while hashing is comparatively “brute force and dumb.” Drake therefore proposed keeping public keys hidden behind hashes and moving assets to a fresh address after every transaction.
- A signature breakthrough would threaten far more than token balances. Doug says forged signatures could enable website and message impersonation, disrupt banking apps, and let attackers spoof validators so proof-of-stake chains become reorgable at will. His stark framing: “The internet as we know it is kind of broken.”
- Doug’s base case is a gradual migration, closer to Y2K remediation than an overnight extinction event. Hash-based signatures already offer a fallback, although they are cumbersome and potentially too large even for a Solana transaction. The decisive variable is speed: years permit coordinated upgrades; weeks might not.
- He is skeptical that OpenAI has secretly cracked discrete logarithms or elliptic curves because AI’s headline math results remain narrow and incremental. Models are “very, very good at checking a box,” often crossing a theoretical barrier by a tiny amount without immediate practical utility. He considers a sudden, operational cryptographic break possible but “pretty unlikely.”
- AI may strengthen crypto security before it undermines cryptography. Formal verification can turn program requirements into theorems and prove that code follows them, attacking the smart-contract bugs Doug considers today’s dominant risk. He thinks this could produce “quote unquote ‘perfect security’” and points to formally verified work in Zcash and his own Lean-based EVM AMM.
- The actionable defense is ordinary operational security, not constant wallet migration. Doug recommends holding most funds in a two-of-three multisig—Squads on Solana or Gnosis Safe on Ethereum—using two hardware wallets with different stacks plus a trusted mobile wallet. “You’re more likely to fuck up and lose your coins” attempting bunker mode than to have AI crack them today.
1. AI crossed from contest math toward open research
Doug traces the progression from SAT questions to AMC and International Math Olympiad problems, then to challenging Erdős problems. He says they were difficult, though not necessarily especially important.
The “shot across the bow,” in his telling, was AI solving Navier–Stokes, one of seven $1 million Millennium Prize problems, only one of which had previously been solved. He stresses that the result was technically significant even though its construction was narrow and had no immediate practical application.
OpenAI then dumped 400 open math problems at once. More important than the count, Doug says, was its breadth: models that had been uneven across subfields suddenly “seem good at, like, everything now.”
He attributes the shift mainly to chain-of-thought reasoning and reinforcement learning on verifiable rewards, not one magical discovery. Base models remain “autocomplete…super, super souped up,” but can now write out reasoning, check their work, and hill-climb on problems with verifiable answers: “data plus scale equals intelligence.”
2. Cryptography’s assumed hardness is suddenly repriced
Modern cryptography depends on operations believed to be difficult, even though their difficulty is not always proven. The prior pace of mathematical progress made hard problems seem unlikely to break soon; discrete logarithms are one example. AI has widened the possible variance over the next year or two.
Doug thinks Justin Drake oversold “months, not years,” yet accepts the underlying warning. Elliptic-curve and lattice signatures have mathematical complexity that might conceal “some weird way to reverse it”; hashing is less complex and comparatively brute-force.
Drake’s proposed bunker mode exploits that distinction: Ethereum addresses hash public keys, but spending reveals the key. His remedy is to move everything into an unused address, then repeat after every transaction; Doug calls that “pretty extreme” and doubts it is warranted today.
3. Broken signatures would become an internet-wide crisis
Doug broadens the blast radius beyond crypto: Zoom conversations, authenticated messages, websites, and banking apps all rely on signatures to establish identity. If attackers could forge them, “anyone can fake being someone else,” potentially forcing people back to in-person banking until systems migrated.
Proof-of-stake chains face a second-order failure even if individual balances remain protected. Forged validator signatures could permit arbitrary reorganizations, meaning “the chain’s not gonna run.” Doug says Bitcoin would be the exception—the chain that keeps running—because proof of work depends on hashing rather than signatures.
Hash-based signatures provide an escape route, and Doug thinks a lot of military systems have moved to them while NIST pushes them. Their drawback is operational: they are “really big” and “a pain in the ass”—so large, he says, that one signature would not fit inside a Solana transaction.
4. The Y2K analogy makes timing the decisive variable
Doug remembers 85%–90% of people expecting Y2K to be fine while roughly 10% hoarded food. The calm outcome did not prove the threat imaginary: “People spent a lot of money to fix all those Y2K bugs.”
His analogous crypto scenario turns on fast versus slow. A breakthrough unfolding over years gives networks time to adopt hash signatures; publication one month before exploitation leaves little runway. A week-scale leap looks “very unlikely,” months remain doubtful, and years are plausible enough to prepare for.
Secret OpenAI breakthroughs cannot be ruled out, but Doug sees no supporting pattern. Existing results typically cross a boundary “by a very, very small amount”; cracking elliptic curves with immediate practical consequences would represent a qualitatively different capability.
Early warning could come from lower-security “canaries”: perhaps $10,000 locked behind a 96-bit construction rather than 256 bits. An attacker improving gradually should harvest that low-hanging fruit before stronger systems—Doug’s signal to say, “Oh, shit, I gotta go into bunker mode.”
5. AI’s nearer-term payoff is better code and practical math
Smart-contract errors remain the risk Doug actually loses sleep over after writing code associated with millions of dollars. Formal verification lets developers encode required behavior as theorems and prove the implementation follows them, making feasible work that was previously too tedious and lengthy for humans.
He cites Zcash’s new shielded pools and his own EVM AMM written purely in Lean as early examples. Shielded Zcash may also be relatively defensive in a doom scenario because an observer does not see an exposed signature, but a Merkle-tree-based shield—though Doug concedes consensus could still break.
Doug tested AI’s research usefulness on integer multiplication after people mocked OpenAI’s improvement as cosmically impractical. He reports OpenAI’s form as roughly
2^1.82m − 1.82and his revised form as2^1.34m—a huge reduction that remains astronomical. The transcript also contains a separate “1 over 2^82” phrase, so the normalized figures are unclear; others reportedly reached2^16, though he said he needed to check.Absent one of those canaries being broken, his threat model remains mundane: North Korean hacking, zero-days on computers, and wallet compromise. His own two-of-three multisig setup diversifies hardware and software failure modes, while Baranos AI—his early project—is exploring AI adjudication of prediction-market outcomes.
Full transcript
Boom. Mr. Doug, how are you, man?
Hey, Trent. How’s it going?
I’m good, dude. It’s good to see you, man.
Yeah, good to see you. Thanks for having me on. I’m excited.
Dude, I appreciate you coming on. I think this is basically our 1-year anniversary of the last time you came on, which I think was 10/11/2025.
Yeah. Yeah.
So—
Every year around this time, something crazy happens in crypto.
Every year around this time—
Why are coins selling off?
I know. I wish it could be on a more bullish day, but the somber days are just as important. Do you want to give a really quick intro into who you are and why you’re a math wizard, by the way? I wasn’t aware of this background.
Yeah, yeah. My name’s Doug. I’ve been in crypto for a while now, working on a perp exchange, Superluminal, that’s not out. Now I’m working on a crypto AI project as well, Verinos AI, which is kind of on-chain LLMs and inference.
Math-wise, I don’t know, I’ve always just been a math hobbyist. I’ve always wished I was a mathematician. Maybe not now, though, because none of them seem to have jobs.
Yeah.
So—
Sick. Well, thank you for being here with us. I guess to start, can you lay out— I mean, you can get niche and complicated if you want, but maybe at a high level as well—what exactly did OpenAI release, and what have they been releasing on the math side of things, and why is it so significant?
1. OpenAI's Mathematical Breakthrough
Yeah. No, that’s a great way to get started. What happened is, obviously, we’ve had these LLMs, these language models, and their capabilities have been increasing over time. Starting this year—or I guess for a little bit—you’ve kind of seen them improve over time.
Even 2 years ago, it was like, “Okay, can they do SAT-type math problems?” Then they started doing these AMC, these kinds of math contests that smart high school kids would do, like IMO, the International Mathematical Olympiad. Those aren’t new problems, but they’re definitely problems that only the brightest kids could get, and they started doing those.
Then, beginning this year, they started solving what are called Erdős problems. They’re named after this mathematician, Erdős, who basically came up with, I don’t know, something crazy, like 1,000 problems or something. These aren’t really that important, but they’re challenging, and any time you solve one, it’s kind of hard to do. So they started solving those this year, and it wasn’t just OpenAI. People were using ChatGPT Pro and solving them.
The real shot across the bow this year was when they solved Navier–Stokes, which is a Millennium Prize problem. Basically, there are these 7 problems that had a prize of $1 million attached to them. At the beginning of the century, they said, “What are the hardest, biggest problems in math?” Seven of them. Here they are. One had been solved up to that point, so the fact that they solved one was huge.
It has to do with fluid and fluid dynamics. It’s complicated, but the idea is that this is a really big accomplishment. Great mathematicians have been working on this for a long time. Obviously, this would be huge if a mathematician solved it.
Then, just—what was it?—2 days ago, they released 400 open problems in math, and these were all pretty significant. I don’t even know if any were on the level of a Millennium Prize problem, but the fact that they just dumped 400 at once was like, holy shit.
The other small thing is that, so far, AI in math has been good at certain areas and not great at other areas. Even if it’s combinatorics or certain subfields, it doesn’t really matter. It has been good at certain things and maybe not so good at other things.
I think with this problem dump, they seem to be good at everything now. So now it’s like, “Okay, this is the moment.” OpenAI—or AI in general—is really just knocking out math problem after math problem, and pretty significant things across the field.
Yeah. And so what changes now, knowing that these LLMs have the capability to solve problems at this level?
Yeah. I mean, obviously, the thing people are talking about in crypto is that there are a lot of things in cryptography, which is obviously the root of cryptocurrency, where we don’t have a great sense of how hard something is. There are certain things we don’t know how hard they are to do, and a lot of cryptography relies on the idea that we think doing one operation is easy and doing another operation is really hard.
But we’re not 100% sure how hard it is. People have been trying for a while to solve things like a discrete logarithm in an easy way, but we don’t know 100% for sure. I think what’s happening now is that we don’t know what the pace of math is going to look like for the next year or 2.
We’ve had a sense of, “Okay, math advances roughly this much per year, so this is a really hard problem. It’s pretty unlikely this is going to get broken anytime soon.” But I think what people are worried about now is that all bets are off the table, and we don’t know how fast things are going to go.
A lot of assumptions that we could make about certain things are now like, “Okay, I’m not really sure.” In the next 6 months, there’s a chance that a lot of things in this recent jump could be things people didn’t expect to be there—very counterintuitive results. So I think people are now thinking about the risk that these long-held assumptions could get overturned very quickly.
Was there any pinpoint advancement in these LLMs that allowed them to all of a sudden be incredible at math?
2. How LLMs Learned To Reason
Yeah. I don’t know if there’s any one thing. The biggest thing was probably chain-of-thought, which is basically the LLM thinking about something and writing it out. It writes out its thinking steps and keeps going back and forth.
That was the big one for reasoning. Before, LLMs were more vibe-ish, and that’s often why you got hallucinations. They’d be like, “I kind of feel like this is right.” Now they actually have to write it out, check their work, and review it.
That, and maybe what they call RLVR, which is reinforcement learning with verifiable rewards. Basically, they take these base-trained models, and the way the base-trained model works is that you just dump a whole bunch of data from the internet into it and run it. So it’s literally autocomplete.
When you think, it’s the same kind of autocomplete you had on your phone 20 years ago, but super, super souped up. Those are the base models. Then they took those and started plugging in real math problems to get results, and they could kind of hill-climb those.
So it’s not one thing, but the data has gotten a lot more accurate, and the labs have invested a lot more in these types of problems, I guess. I think the story of LLMs is just: data plus scale equals intelligence.
And so then Justin Drake makes this post, which is like a bomb going off on crypto Twitter. Could you maybe break down what he is suggesting could happen—not fear-mongering, but what he’s suggesting—and what your personal thoughts are on the Justin post?
3. Cryptography Faces New Uncertainty
Yeah. I think it’s probably oversold, what he’s saying, but there is definitely something worth breaking down and engaging with there. Justin’s point is—and going back to what I said before—a lot of things in cryptography are based on things we don’t know for sure, but we’re pretty sure they’re hard. We don’t know if they’re hard.
Also, like I was saying before, we don’t know the pace of advancement right now. Things that we think are hard may actually turn out not to be hard, and we might discover that pretty soon. I think Justin’s post was saying months, not years. I think that’s probably a hard thing to say, but the point is that there’s a wider variance now.
The other point, which I think is a little more subtle, is that certain things in cryptography are more complex than other things. Pretty much all crypto chains today rely on elliptic-curve signatures. Long story short, that just means that we have these curves, and you have to find 2 points on the curve, and those are hard to do.
And then people are also looking at lattice signatures. Again, these are complex things. Justin's point was that the more complex things are probably more likely to have some weird way to reverse them.
Got it.
Versus hash-based signatures, like hashing. That's the other thing: hashing is just brute force and dumb, so there's probably not a weird backdoor into it. Justin's point specifically was that you should not have your public keys exposed anywhere. You should only have the hash of the public key, because reversing a hash is probably a lot harder than going from a public key to a private key.
With Ethereum, your address is actually a hash of your public key, but when you send a transaction, you have to send your public key so someone can see it. His point was that you should go into bunker mode and move all your assets into an address that's never been used before. Every time you use it, move everything to a virgin address and just keep doing it. That seems pretty extreme. I don't know if we have to worry about that today, and if we did have to worry about it, I think enough things would be fucked that your ETH probably wouldn't be worth anything—
So this is a tough question, right? It's like people are, you know, if crypto— We have bigger problems than your coins going to zero. I'm like, personally, that is my biggest problem. I don't know about you all, but could you touch on that a little bit—cryptography broadly? What is he saying could happen, and if it were to happen, what is it connected to? What are the implications beyond just Bitcoin and crypto?
reserved anyway.
4. Signature Failure Breaks Everything
Yeah, everything relies on it. Even the fact that we're talking over Zoom and people can't eavesdrop on our conversation, or they can't fake a conversation, or if you send me a message, I know it's from you and not from North Korea—those all rely on pretty much elliptic curve signatures, these standard signatures.
If anyone can fake a signature, the internet as we know it is kind of broken, because anyone can fake being someone else and anyone can fake a website—your banking website. So maybe the whole internet's broken and you just have to show up at your bank for a little bit. You can't use your banking app. That happens.
The other thing is, if signatures are broken, pretty much every proof-of-stake blockchain is affected. Even if your coins are safe, Ethereum relies on signatures from the validators. So if somebody can lie about the signatures of the validators, anyone can reorganize the chain at any time, and the chain's not going to run. The only chain that would actually be running is Bitcoin, because it's proof of work and relies on hashing, not signatures.
How would this play out from this scenario where it's like, okay, if anyone could fake a signature, then the entire internet as we know it ceases to exist? Obviously, there are massive, massive, massive implications to that. There are bank accounts. There's every message that's ever been written everywhere online. How do you even—playing what if—how do you see this playing out from here? This is government intervention stepping in. How does this—
The good thing is that we do have these things called hash-based signatures, which are probably not—again, anything could be breakable, but these are just dumber, so there are fewer hooks into them to break them. We do have these things called hash-based signatures. What we would probably have to do is move everything to hash-based signatures.
The problem with hash-based signatures is that they're really big and annoying to work with. You could even have a blockchain with hash-based signatures, but they're so big that you couldn't even fit one signature into a Solana transaction. That's why nobody uses them. They're just a pain in the ass. But they exist, so you could use them. You would eventually start moving more and more things over.
What uses them right now?
I think a lot of military stuff has moved to them. I know NIST is pushing them pretty hard. Blockchains don't use them because we care about performance. But maybe we should start thinking about it.
Interesting. It feels like a crazy generational fearmonger, because it's like this thing could hypothetically happen in the future at a rate of acceleration that we can't measure, and there's no way to disprove that it can't happen.
There's no precedent for it, is the problem. I guess I'm tempted—I'm old enough that I remember the Y2K stuff.
Okay.
Yeah.
People are saying, "I wasn't even born during Y2K." But were people really—
Yeah. Zoomers don't get that. Y2K was kind of like this, right? Most people—85% or 90% of people—were like, "Yeah, Y2K's probably fine." Ten percent of people were fucking nuts, like, "Okay, I'm hoarding food and stuff. I don't know what's going to happen."
Obviously, it didn't happen. But on the other hand, people spent a lot of money to fix all those Y2K bugs. It's not like everyone was just like, "Oh, the worries were unfounded." They were there, and you had to do a lot.
I guess I'd say with this AI stuff, there's a big difference between this being super fast and this happening over months. If the AI goes crazy fast, which I think is very unlikely—as someone who's using AI to try to solve math problems, I don't think it's going to solve anything in a week. I doubt it's going to solve any of these big things even in months. Maybe in years, but if you have years, you have time to switch stuff over.
If it's like, "Oh, shit, we're publishing something and it's going to be out in a month," that goes faster. So it's fast versus slow. I think so far, though, what we've seen from the results is that AI is very, very good at checking a box. Some of these results with OpenAI technically broke some barrier, but they broke it by a very, very small amount.
I think AI is very good at that, and that ends up not being that practical. Then you have to go back and forth
Got it.
to get anything practical. That's why I'd be a little bit more optimistic, at least in this scenario. These things are smart, but they do incremental things more than I think people realize.
What do you think about this idea that there might be some secret breakthroughs that OpenAI has already solved internally? How likely is that?
You can't rule anything out. I think that's pretty unlikely.
Really?
Well, again, these would be pretty big breakthroughs. If they had a breakthrough on discrete log, elliptic curves, or anything like that, it would be bigger than anything we've seen. Even now, Navier–Stokes is another example. They technically checked the box for the Millennium Prize, but there are no practical applications for it yet because the way they did it is very, very specific.
So far, we haven't really seen any major mathematical result from AI that has immediate practical applications. If they broke cryptography this way, it would be very different. I think it would imply even more capabilities than what we've seen from AI.
Okay. I like the framing of that. What about on the positive side? Are there any positive implications from AI's newfound capabilities with math?
5. AI Could Deliver Perfect Security
Yeah, I think enormous ones. The story here is that, especially for someone who's written smart contracts and has had millions of dollars in smart contracts I've written, that's fucking terrifying. Going to bed and thinking, "Oh, shit, I really hope I didn't make any mistakes in the code I wrote"—that's definitely one of the most stressful things about being a crypto developer.
One cool thing we can do now is what's called formal verification, where we can write programs, structure them as theorems, and have proof systems prove that this code follows this. You have to get your conditions right, but the whole idea is formal verification. They're doing this with Zcash now, with the new shielded pools.
That's just huge from a security standpoint. So far in crypto, the security risk isn't cryptography breaking; it's smart-contract hacks. I'm super excited that AI can do formal verification. Writing formal verification before was just so tedious and so long that humans couldn't do it.
We're already doing stuff now where I actually put a little AMM contract out on EVM, which is purely in Lean, so formally verified—all of that stuff. From a security perspective, I think it is possible we get quote-unquote “perfect security” because of AI.
Perfect security because of AI? That's insane. Is Zcash uniquely positioned for some of these doom scenarios?
I think Zcash is actually, for a doom scenario, maybe better situated, especially if you have shielded Zcash. That's probably one of your better situations, because shielding gives you two layers. One is your signature, but the other is that technically nobody knows your signature, and it's this Merkle tree.
Yeah.
If you have to go into bunker mode, like Justin Drake says, I'm going to say Zcash is guaranteed. I mean, consensus would break too, but Zcash is probably one of the better places you can go to bunker mode.
Okay, I'll take that. What are you working on right now on the math side of things? Your timeline has been electric the last 72 hours.
6. AI Finds Practical Math Uses
As soon as OpenAI dumped those, I did these little stupid explainers where I say, “Okay, try to take a thing and put it into a little comic or whatever,” and people kind of like them. So I did that, and then I thought, “Okay, this is kind of fun.”
Then I asked it, “Okay, out of any of these problems, if I just wanted to make an improvement on what OpenAI published, which one do you think would be fun?” It was this one on integer multiplication, which is basically: You have 2 really, really big integers, and how hard is it to multiply them together? That's basic asymptotic complexity, which just means, as they get bigger, how much harder does it get?
OpenAI's result was really cool because it found that what we thought was the limit wasn't quite. We could go below the limit, but only very, very slightly below the limit. Everyone was clowning on it, saying, “What's the point of this? What's the point of this, Doug? It doesn't even matter. There's more atoms in the universe.”
I also thought it would be kind of cool to show that you have to be more optimistic about math. Things will advance. So I took this problem and put it into Astra. I was going back and forth. I like to think I kind of contributed, but it was a lot of Codex.
Basically, the rate of improvement was 2^1.82m − 1.82, so 1 over 2^82. We took it down to 2^1.34m—so a quadrillion, quadrillion, quadrillion. A huge reduction. It started astronomical. It's still pretty big, but not as astronomical.
Then people took off with it, and now people are doing it. Somebody has it down to 2^16, I think. I have to go back and check everything now. So it's actually getting to practical applications, which was kind of my hope. I want people to be less cynical about this stuff.
Whoa.
If we could actually get it to something practical, where people are clowning on it as, “What is even the point of this?” I think it would be good for society.
How many tokens to get there?
I burned through my Codex subscription, and then I made my wife give me her Codex tokens. I might have to do some resets.
That's not too bad.
Yeah, not too bad.
That was optimistic. I like that. I guess one more on the original doomsday: Are there any sort of flare-gun signals to look for—“Oh, shit, we maybe have a bigger problem than we thought”?
7. Cryptographic Canaries Signal Danger
I think there are canaries. One is, how many bits of an address? Because what is Ethereum—160 bits? I forget how many. In all these signature schemes, you can put more bits into them. Maybe it's 256 bits. Basically every signature scheme uses a lot of bits, and those numbers of bits are hard to break.
I think there are canary amounts out there where it's like, okay, a certain amount—$10,000—in a smart contract that's 96 bits instead of 256 bits.
Yeah.
Those are kind of interesting, right? Because if someone's ramping up and getting good at it, they're going to hit the low-hanging fruit first.
Yeah.
If you see someone break those types of canaries—and again, I don't know the specifics off the top of my head; I should—but there are canaries out there that should get hit before everything else. Then you're like, “Oh, shit, I gotta go into bunker mode.”
Okay. This is basically my last question for you on the doom stuff. You said what Justin proposes is a little bit extreme. Bunker mode even sounds a little bit crazy. Is there any precaution you think people should take right now that's more practical or reasonable?
I think doing—
Are you doing anything?
Yeah. I think doing bunker mode is pretty extreme. You're more likely to fuck up and lose your coins if you try to move them every single time than actually have the AI hack them, especially today.
I would say the biggest risk still isn't cryptography; it's North Korea. North Korea is getting way better at hacking, and you don't know what kind of zero-days are on your computer. For the money you have, use a multisig: Squads on Solana, Gnosis Safe on Ethereum.
Put most of your money in a multisig, 2 out of 3. I would use 2 different hardware wallets, so different stacks, and maybe 1 mobile wallet on your phone that's pretty well trusted. If 1 stack gets hacked, you're okay. Then you have this recoverability thing.
That's what I personally do, so I think that's kind of the best option right now.
I feel like we're due for a security stream at some point. That was good. That was really good.
Doug, you're a legend, man. Honestly, I appreciate you coming on here optimistic. I figured you would be, but I wasn't sure what the vibe was going to be.
Do you have any, like, sort of, one, sign-off that you want to shill, and then two, anything you think people should read, pay attention to, anything, like, you should want to direct people to? Um, yeah.
Uh, I guess I'll, I guess I'll shill, like, the AI stuff we're doing right now, so it's Baranos AI, B-A-R-A-N-O-S AI. Um, we're early. We're just doing early stuff. But I think, like, we have some cool stuff coming there with, like, prediction markets and using, like, AI to, like, judge, like, prediction market results and stuff. So that's, uh, that's my shill.
Doug, you're awesome, man. Look, it's our annual stream. Hopefully next one will be—
Yeah. Hopefully next one will be in bunker mode.
ATHs.
Next year, 2027.
Yeah, not bunker mode, but, dude, I appreciate you taking some time out to come on. Much love, man.
Awesome.
Good to talk to you.
Thanks, man. Appreciate it.
Have a good one. Peace.