Swan's Klippsten Says AI Threatens Bitcoin More Than Quantum. He's Half Right.

A kid with a $20 chatbot subscription can now read code faster than an exhausted auditor on a Friday night. That's the short version of why Cory Klippsten, the founder of Swan Bitcoin, says artificial intelligence keeps him up at night more than quantum computers do.

And he's largely right about the direction. Maybe a little off on the urgency. We'll get there.

The near threat versus the far one

Klippsten's argument, made publicly this week, is that attackers are already using large language models to hunt for security holes. A handful of prompts, a pasted contract, a wallet implementation, and the model spits back a plausible attack path. No PhD required. No months of manual review. Just a curious person with an internet connection and a model that never gets bored.

Quantum computing, by contrast, is a physics problem with a long runway. Google's Willow chip turned heads in December 2024 with 105 qubits, and the headlines practically wrote themselves. But breaking the elliptic curve cryptography that secures Bitcoin wallets, secp256k1 to be precise, would take millions of physical qubits working with error correction that nobody has built yet. Best estimates put that at a decade out. Probably more.

So the comparison isn't really AI versus quantum. It's a threat that exists today versus a threat that exists in a whiteboard sketch.

Granted, quantum fears sell better. There's something cinematic about it, a warehouse-sized machine cracking private keys while the price chart goes vertical. AI hunting for a bug in your multisig setup is less fun to imagine, mostly because it's a Tuesday.

Cryptography isn't the soft spot

Here's where I think Klippsten is pointing at something the industry has been slow to internalize. Bitcoin's math is fine. The math has held for 16 years and it'll hold for a while longer. The failures keep happening one layer up, in the humans and the tooling around the humans.

Look at February 2025. Roughly $1.46 billion walked out of Bybit in the largest crypto theft on record. The cryptography didn't break. A developer's machine got compromised, a Safe wallet interface got swapped, and a signer approved what they thought was a routine transaction. Social engineering, lubricated by software.

Now pour a language model on top of that. It can read a target's entire public footprint, their GitHub commits, their conference talks, their job history, their tone of voice in old forum posts, and draft a message that sounds exactly like their boss. In seconds. Thousands of variants. The skill floor for a convincing spear-phish just dropped through the basement.

That's the asymmetry people miss. Defense teams get the same models, sure. But attackers only need one hit, and the cost of a miss is zero. Defenders need to catch everything, forever, and they're already tired.

The question worth asking: if a model can audit your code as well as an outside firm charging $200,000 a pop, why are so few teams running one before they ship?

Cui bono

There's a market angle here too, and I don't think it's getting enough attention.

Quantum FUD has quietly become a product. Tokens, vaults, post-quantum chains, all pitched off a threat that won't be actionable for years. It's a clean narrative because nobody can disprove a future attack. AI-assisted exploitation is messier. It's happening now, it's hard to package as a ticker symbol, and it mostly punishes small teams and solo builders who can't afford a security department.

So the people who lose are the little guys. The people who win are the ones selling detection, monitoring, and code review that's augmented by machines. Admittedly, that's a less exciting trade to talk about at a conference.

Color me skeptical, but I don't think quantum is the threat the industry should be budgeting against first. I've watched the same slide deck about Shor's algorithm for three years. Meanwhile, the actual losses keep coming from stolen keys, phished signers, and a compromised laptop in someone's home office.

If you're holding your own coins right now, your threat model shouldn't assume an attacker with a million-qubit machine. It should assume an attacker who has read everything you've ever posted and can write to you in your own voice.

The takeaway

Klippsten's thesis is simpler than the headlines make it sound. Bitcoin's cryptography is the one part of this system that's been stress-tested for over a decade and hasn't flinched. The people around it haven't been tested nearly as hard.

So, concrete advice. Move to multisig if you haven't. Verify addresses out of band, over a phone call, with a human voice, not a chat window. Treat any urgent message about your holdings as hostile until proven otherwise. And if you're building, assume your code has already been read by something that never sleeps and doesn't need a coffee break.

The prompt is the new crowbar. Admittedly, that's less romantic than a quantum computer. It's also a lot more likely to be the thing that actually gets used on you this year.