Drake said the math behind bitcoin and ether wallet signatures, known as elliptic curves, follows tidy patterns that a powerful enough AI could learn to exploit. Hash functions, which turn any piece of data into a fixed-length digital fingerprint that can’t be worked backward, are built to scramble information with as little pattern as possible.
AI-assisted attacks on crypto systems have already cost real money.
Last December, Anthropic researchers showed frontier models could write working exploits against simulated copies of real DeFi contracts. In late July, a volunteer group called the Bitcoin Red Team used AI models to sweep 390 Bitcoin software projects in about 27 hours, logging nearly 5,000 possible flaws, 85 of them rated critical.
On July 30, an attacker began draining Coldcard hardware wallets through a five-year-old firmware bug, taking at least 1,367 BTC, and maker Coinkite said it suspected AI helped find the flaw.
Days later, BTCPay Server confirmed attackers had stolen funds from merchants’ Lightning nodes through a flaw first surfaced in an AI-assisted audit, and on Aug. 27, Core Lightning’s developers issued an emergency warning after AI-generated bug reports turned up real vulnerabilities in their software.
Researchers also used AI coding agents to improve a calculation inside a future quantum attack, as CoinDesk reported in September, though that work still required quantum hardware and covered only part of the attack.
