The AI threat to bitcoin is real, and it is not aimed at the cryptography

An investor put a number on it. Liron Shapira gave 50% confidence that bitcoin loses more than half its value within two years, because AI would weaken what users believe about the network's security. Vitalik Buterin took the opposite side, and the distinction he drew is the useful part of the exchange.

Quote source: crypto.news, 7 September 2026

The distinction Buterin drew

Buterin separated two things that headlines keep merging. One is a break in bitcoin's cryptography: a practical weakness in SHA-256 or in proof-of-work that lets an attacker produce valid hashes far more cheaply. He put the probability of that at tiny.

The other is everything around the cryptography. Node software, wallet code, mining pool systems, exchanges, and the internet infrastructure connecting them. AI helps attackers find flaws in all of it faster, and it helps defenders review code and detect intrusions faster too. His argument is that client developers and pools can answer those attacks with ordinary upgrades, which is far easier than an emergency change to bitcoin's monetary rules.

Our own year supports him

The three largest losses we have covered since July all started outside the cryptography.

This afternoon we published the Liquid Network incident: about 4,000 BTC left a federation wallet because of a rangeproof cache bug in the Elements software, and the fix had already been merged to the repository. Last week a shipping contractor leaked the home addresses of Trezor buyers. In July, Bitcoin Optech reported that some COLDCARD-generated wallets had insufficient entropy in key generation.

None of those touched SHA-256. Each of them cost real money.

What the bet is not

Outlets reporting the exchange noted that Buterin backs his view with about 90% of his net worth. He holds that position already, so it is an existing exposure rather than a wager anyone can settle. Nobody has escrowed anything, and no date resolves it.

His word tiny is an assessment, not a measurement, and it answers a narrow question. It says a cryptographic break is unlikely. It does not say bitcoin holders are safe, and readers collapse those two sentences into one.

Shapira's forecast has the mirror problem. It is about price and belief rather than about a technical event, so it can come true through a market panic that never involves a broken hash. Both men can be right at once, which is why the framing as a bet flatters the disagreement.

The thing most likely to take your coins in the next two years is a bug in software written by people, found faster than before by a machine. That is what happened at Liquid, and the patch for it existed days early.

None of this should be read as personalized investment advice.