Drake pointed out that the mathematical principles governing bitcoin and ether wallet signatures, which utilize elliptic curves, exhibit orderly trends that a sufficiently advanced AI could potentially manipulate. Hash functions are designed to transform data into a fixed-length digital signature that is resistant to reverse engineering, effectively jumbled to minimize recognizable patterns.
AI-driven assaults on cryptocurrency systems have already resulted in significant financial losses.
In December of last year, researchers from Anthropic demonstrated that advanced models were capable of creating effective exploits targeting simulated versions of actual DeFi contracts. In late July, a collective known as the Bitcoin Red Team employed AI technologies to thoroughly examine 390 Bitcoin software projects within approximately 27 hours, identifying close to 5,000 potential vulnerabilities, with 85 classified as critical.
On July 30, an assailant started extracting funds from Coldcard hardware wallets by exploiting a bug in the firmware that had existed for five years, successfully stealing at least 1,367 BTC. The creator, Coinkite, indicated that they suspected AI played a role in discovering this vulnerability.
A few days later, BTCPay Server revealed that funds were pilfered from merchants’ Lightning nodes due to a vulnerability initially identified during an AI-assisted assessment. On August 27, developers of Core Lightning issued an urgent alert after AI-generated bug reports revealed genuine weaknesses in their software.
Additionally, researchers utilized AI coding agents to enhance calculations related to a prospective quantum assault, as reported by CoinDesk in September, although this endeavor still necessitated quantum hardware and only addressed a portion of the overall attack.