Bitcoin’s Security Faces a 76% Risk by 2046, AI Model Predicts
AI models predict Bitcoin’s signature schemes could face a 76% chance of being broken by 2046, driven mainly by quantum computing threats.
2 min
Written with AI assistance from the reporting linked below, and reviewed before publication. How we write
Bitcoin.com News on Oct. 11, 2026 published the results of an experiment that queried four large‑language models for probability estimates that Bitcoin's ECDSA and Schnorr signatures over secp256k1 could be practically broken. The study examined the likelihood that an attacker could derive private keys from exposed public keys at scale across four time horizons.
The experiment was prompted by an X post from onchain analyst Willy Woo, who shared figures attributed to Opus 5.5. Those figures showed a 0.5% quantum risk and a 0.8% classical‑or‑AI risk within two years. Bitcoin.com News expanded the prompt to two, five, ten, and twenty years and collected responses from Deepseek Deepthink, ChatGPT‑6 Astra, Grok Expert, and Claude Opus 5.5.
AI Models Diverge on 20‑Year Risk
At the twenty‑year horizon the models produced a wide spread. Deepseek Deepthink assigned a 34% cumulative probability of a practical break. ChatGPT‑6 Astra estimated approximately 76%. Grok Expert gave 60%, and Claude Opus 5.5 gave about 66%.
At ten years the range narrowed but remained material. Deepseek reported 6%, ChatGPT‑6 Astra about 36%, Grok Expert 30%, and Claude Opus 5.5 roughly 31%. At five years the estimates clustered between 1% (Deepseek) and about 10% (ChatGPT‑6 Astra and Grok Expert). At two years all models placed the combined risk at or below 2%.
Quantum Computing Seen as Primary Threat
Across every model and every horizon beyond the near term quantum computing accounted for the overwhelming share of the projected risk. The classical‑or‑AI column stayed in single digits even at twenty years: Deepseek 5%, ChatGPT‑6 Astra about 2%, Grok Expert 10%, and Claude Opus 5.5 2%.
The models’ follow‑up explanations, limited to 100 words each, converged on the same reasoning. No cryptographically relevant quantum computer exists today. Current systems remain far below the logical‑qubit count and error‑corrected scale required for Shor's algorithm. Expert surveys assign meaningful likelihood to such machines appearing in the mid‑to‑late 2030s or later.
Classical and AI‑Assisted Risk Remains Low
None of the four models identified a known practical classical break against secp256k1. The best‑known attacks remain fully exponential, and no AI‑driven shortcut has been demonstrated. Deepseek Deepthink and Claude Opus 5.5 kept the twenty‑year classical‑or‑AI probability at or below 5% and 2% respectively. ChatGPT‑6 Astra placed it at about 2%, while Grok Expert was the outlier at 10%, still an order of magnitude below its own quantum estimate for the same horizon.
Model Reasoning Highlights Consensus Gaps
The follow‑up responses revealed where the models agree and where they differ. All four cited the absence of fault‑tolerant quantum hardware at scale as the reason for low near‑term probabilities. All four referenced expert surveys that assign growing likelihood to cryptographically relevant quantum computers in the 2030s. The divergence at twenty years appears to reflect different weightings of engineering‑progress uncertainty versus the length of time available for barriers to fall.
Bitcoin.com News notes that all figures are speculative model‑generated estimates, not verified probabilities. Watch for the next round of expert surveys on quantum timelines, any published reduction in the logical‑qubit requirement for Shor's algorithm on secp256k1, and whether Bitcoin developers advance a post‑quantum signature proposal from research to testnet deployment.