Willy Woo's Bitcoin (BTC) Post Triggers AI Estimates of 76% Crack Risk by 2046

Willy Woo Ignites an AI Forecast Debate

On Sunday, October 11, 2026, on-chain analyst Willy Woo shared AI-generated estimates on X putting the combined odds that the digital signatures securing Bitcoin (BTC) could be broken within two years at 1.3%, a post that set off a new round of debate about the network's long-term security. The figures Woo passed along came from Claude Opus 5.5, which put the chance of a quantum computer cracking the signatures inside two years at 0.5% and the chance of classical mathematics or AI-assisted cryptanalysis doing the same at 0.8%. Neither figure is a disclosed vulnerability; both are model judgments, and that distinction frames everything that follows. The debate concerns the signature layer rather than the Bitcoin (BTC) price, though the attention it drew reflects how much value sits behind those signatures. Building on the question Woo raised, the test was then expanded to roughly four AI models across four horizons: two, five, ten and twenty years. Each model was cast as a mathematician, quantum computing specialist and on-chain analyst, and asked to estimate the probability that Bitcoin's current scheme, ECDSA and Schnorr signatures over the secp256k1 curve, could be practically broken: an attacker deriving private keys at scale from exposed public keys, either with a cryptographically relevant quantum computer or through classical or AI-assisted advances in cryptanalysis. Signatures are the data the Segregated Witness upgrade moved out of the transaction body, so this layer is where a break would bite. The models were told to anchor on current expert consensus, which holds that practical quantum risk sits well beyond 2030, and on the absence of any known practical classical break of secp256k1, rather than on speculative worst-case scenarios. Answers were to be short lines, percentages only, split into a quantum figure, a classical or AI figure, and a combined total for the Proof of Work network's signature scheme.

@willywoo · X post

Shared AI-generated estimates on X.

View on X

Where Four Models Diverge by 2046

Twenty years out, the four models separate sharply. Deepseek Deepthink, the most conservative, put the cumulative probability of a practical break at 0% over two years, 1% over five, 6% over ten and 34% over twenty. ChatGPT-6 Astra sits at the other end: roughly 1% at two years, 10% at five, 36% at ten and 76% at twenty. Grok Expert returns 2%, 10%, 30% and 60% across the same horizons, while Claude Opus 5.5 gives about 1%, 5.5%, 31% and 66%. Every model, whatever its level, treats quantum computing as the dominant long-horizon threat and classical or AI-assisted cryptanalysis as the smaller one. Stripped of the quantum component, the twenty-year classical-only risk collapses to about 5% on Deepseek's reading, roughly 2% on both ChatGPT-6 Astra and Claude Opus 5.5, and 10% on Grok Expert's. The models' stated reasoning follows the same line. ChatGPT-6 Astra said its short-horizon numbers are low because a real attack needs large-scale fault-tolerant quantum hardware, its five-year figure allows for error-correction and hardware progress while major engineering barriers remain, and its twenty-year figure simply gives those barriers more time to fall. Its ten-year estimate reflects expert surveys that assign a meaningful probability to cryptographically relevant quantum machines in that window. Grok Expert pointed to expert consensus placing practical quantum attacks on secp256k1 after 2030, noting that current machines remain far below the logical-qubit and error-correction scale Shor's algorithm requires, and that the best known classical attacks against the curve still grow exponentially, with no verified AI-powered shortcut. Deepseek's notes stressed the same absence of any known scalable attack today, which is why its two-year floor sits at zero.

Judgment Estimates, Not Measured Probabilities

The sharpest takeaway from the exercise is what it does not claim: the models themselves describe these as judgment-based estimates, not consensus probabilities of an actual large-scale attack, and no scalable break of the curve is known at publication time. Quantum risk is not new to Bitcoin (BTC) commentary, but the AI-model framing sharpens where the exposure would sit. It concentrates on balances whose public keys have already been revealed on-chain, a profile that skews toward older and larger HODL stashes, while addresses that expose a public key only at spend time keep a narrower window. Note also that the estimates target signatures, not the hash functions securing mining, so even a signature break would not by itself let attackers rewrite blocks. The numbers that would break the alarm are the models' own two-year floors, which run from 0% to 2%; any move above them within this decade would require a concrete hardware or cryptanalysis announcement, and none exists today.