Amid a rise in hacks in the crypto market, a group of developers deployed several AI models to find holes in the code of Bitcoin ecosystem projects. As a result, hundreds of vulnerabilities were discovered.
The hack of Coldcard hardware wallets, which led to user losses exceeding $100 million, triggered an unprecedented review of code in the Bitcoin ecosystem. A group of enthusiasts, informally named Bitcoin Red Team, initiated an audit using advanced AI models and reported preliminary results. In just over a day, they found hundreds of high-severity vulnerabilities and dozens of critical ones.
According to preliminary results, in less than 28 hours of work, the team processed nearly 400 open-source projects. They have already found 85 critical errors and 635 high-severity issues. The announcement did not specify which projects these are or what impact the discovered vulnerabilities could have.
"We reached a rate of 2.31 high and critical findings per person per hour," Bitcoin Magazine quotes the developers.
The developers attracted an undisclosed amount of funding to pay for the computing power of popular AI models, including Kimi K3 from Moonshot, GPT Sol from OpenAI, Fable and Opus from Anthropic, and GLM5.2 from Zhipu AI. It is reported that $40,000 was spent on the computing power of these models, directed at auditing over 390 open repositories in the Bitcoin ecosystem.
For the analysis, a special tool was used, not yet disclosed, designed to identify and reproduce vulnerabilities in critical libraries. In the future, Red Team plans to open-source this tool so that "every Bitcoin company can direct the same AI power at its proprietary code before the next exploit happens."
It was also reported that access to OpenAI and Anthropic models was initially restricted, forcing the team to rely on Chinese open-source solutions. The project's funding, according to Bitcoin Magazine, is provided by the non-profit organization OpenSats, which supports the development of open-source software for Bitcoin.
