A large-scale security audit of the Bitcoin ecosystem, initiated by the volunteer group Bitcoin Red Team, has yielded its first results, which cannot be called encouraging. In the first 29.8 hours of work, researchers recorded 4,962 potential issues across 390 projects related to the first cryptocurrency. Of these, 85 findings were classified as critical, and 635 as high-risk.
So far, the team has managed to reproduce and confirm only 21.4% of the discovered issues. The remaining results require additional verification before they can be considered real vulnerabilities. This campaign was a response to a series of attacks on Coldcard hardware wallets, which resulted in attackers stealing over $100 million.
What exactly Bitcoin Red Team is checking
The objects of analysis include cryptographic libraries, wallets, infrastructure software, and other open-source projects. The initiative involves 16 specialists, including the developer of the Cashu protocol (known under the pseudonym Calle) and AnchorWatch CEO Rob Hamilton.
The work methodology combines the use of AI models with manual verification. Participants have created a number of testing tools that automatically load code, identify potentially dangerous areas, and help simulate possible attacks in a local environment. This approach, known as red teaming, is aimed not just at finding errors but also at assessing the possibility of their practical exploitation to compromise a product.
According to Calle, on average each participant finds about one critical exploit per hour, although he does not disclose the exact counting methodology. The team does not publish the list of affected projects: critical reports are first verified in practice and then passed to developers privately.
Reason for the launch: attacks on Coldcard
The trigger for such a large-scale audit was the exploitation of a vulnerability in Coldcard hardware wallets. According to my analysis of the data, at least 15 different attackers took advantage of this flaw. Confirmed losses from three waves of attacks are estimated at $100 million, and taking into account a suspected fourth wave, this amount could rise to $130 million.
The device manufacturer Coinkite explained that the problem arose after firmware changes in 2021. In some cases, devices used a predictable software-based MicroPython random number generator instead of a hardware one. This led to a reduction in the effective entropy of seed phrases from 128 bits to approximately 40 bits on the Mk2 and Mk3 models, and to 72 bits on the Mk4, Mk5, and Q. The company has released patched firmware, but it does not protect already created seed phrases — owners are advised to generate new ones on updated devices.
AI accelerates the search but creates new problems
Hamilton noted that the main limitation of the campaign is not finding vulnerabilities but coordinating and delivering reports to the right developers. So far, the team has conducted more than a dozen disclosures, scanning 150 repositories.
My comment: This situation highlights the dual nature of automated vulnerability hunting. On the one hand, AI allows finding issues on a scale unattainable for humans. On the other, it creates a huge burden on developers who are forced to process thousands of reports. In the long term, we will need new standards for coordination and prioritization, otherwise the ecosystem risks drowning in its own "bug stream."