$2 AI Test Cracks $100M Bitcoin Heist Vulnerability

A recent experiment using a $2 AI test revealed the vulnerability behind a $100 million Bitcoin theft. The model, GLM 5.2, was able to rediscover the weakness in approximately 20 minutes at a cost of about $2.

The researchers involved in the experiment, led by Dragonfly managing partner Haseeb Qureshi, emphasized that the test did not establish that AI was used in the original theft. Instead, it highlighted how easily attackers and developers can now examine code protecting large amounts of cryptocurrency using low-cost models.

Qureshi noted that traditional vulnerability research required specialist knowledge and many hours of manual work, but AI allows for a much broader initial review of code. This makes it easier to identify weaknesses and assign multiple models to the same codebase to investigate different failure paths.

The experiment's findings have significant implications for the security of cryptocurrency wallets and exchanges. It underscores the importance of regularly reviewing old wallet code using improved AI models, especially when major updates occur. This process can also create opportunities for scammers, so users must be vigilant and never provide sensitive information to unauthorized support representatives or websites.