A large-scale security audit initiated by the volunteer group Bitcoin Red Team has revealed an alarming picture of the state of affairs in the ecosystem of the first cryptocurrency. In the first 29.8 hours of work, auditors recorded 4,962 potential issues in 390 bitcoin-related projects. Of these, 85 were classified as critical, and another 635 as high-risk.

So far, the team has managed to reproduce and confirm only 21.4% of the findings. This means the remaining results require additional verification and cannot yet be considered confirmed vulnerabilities. The trigger for such a large-scale audit was the recent attacks on Coldcard hardware wallets, which resulted in attackers stealing over $100 million.

Focus on open source and AI

Bitcoin Red Team, which includes 16 specialists, including Cashu protocol developer (Calle) and AnchorWatch CEO Rob Hamilton, has focused on analyzing cryptographic libraries, wallets, infrastructure software, and other open-source projects. The methodology is based on a combination of AI models and manual review. The specialists developed tools for automatically loading code, identifying potentially dangerous sections, and simulating attacks in a local environment.

This approach, known as red teaming, allows not only finding errors but also assessing the possibility of their practical use to compromise a product. According to Calle, researchers on average discover about one critical exploit per hour per person, although the exact counting methodology is not disclosed.

Context: lessons from Coldcard

The attacks on Coldcard, which triggered this campaign, revealed a serious flaw in the firmware that appeared back in 2021. Instead of a hardware random number generator, the devices in some cases relied on a predictable software generator from MicroPython. This led to a reduction in the effective entropy of seed phrases from 128 bits to approximately 40 bits on Mk2 and Mk3 models, and to 72 bits on Mk4, Mk5, and Q. The manufacturer, Coinkite, has already released fixes, but they do not protect already created seed phrases, so owners are strongly advised to generate new keys on updated devices.

The problem is not in finding, but in coordination

As Rob Hamilton noted, the main bottleneck of the campaign was not the technical part, but coordination and delivery of reports to developers. So far, the team has spent about $20,000 on various services and made more than a dozen disclosures, scanning 150 repositories. The list of affected projects is not disclosed, and critical reports are first reproduced using a working attack example and only then privately passed to developers.

This incident highlights the growing role of AI in cybersecurity, but also its limitations. Automation speeds up the search but does not solve the problem of verification and communication. The bug bounty market, like development in general, faces a new challenge: how to effectively process the flow of findings generated by AI without creating chaos and overloading development teams. In my understanding, the future lies in hybrid systems where AI will not only find but also help in prioritization and automation of the first stages of interaction with developers.