On August 5, the volunteer initiative Bitcoin Red Team presented the results of the first phase of a large-scale security audit of the first cryptocurrency's ecosystem. In the first 29.8 hours of work, researchers recorded 4,962 potential issues in 390 bitcoin-related projects. Of these, 85 findings were classified as critical risk level, and 635 as high. At the same time, only 21.4% of the discovered issues could be reproduced, which highlights the complexity of verification under rapid scanning conditions.

Who conducts the audit and how

The Bitcoin Red Team consists of 16 specialists working around the clock in a globally distributed format. Among the participants are Cashu protocol developer under the pseudonym Calle and AnchorWatch CEO Rob Hamilton. The methodology includes a hybrid approach: AI models are used for initial code analysis, identifying dangerous areas, and generating attack scenarios, while manual verification is used to confirm findings. This approach, known as red teaming, simulates the actions of a real attacker, allowing an assessment not only of the presence of errors but also of their practical exploitability.

According to Calle, the detection rate is approximately one critical exploit per hour per person, although the exact calculation methodology is not disclosed. The team does not publish a list of affected projects: critical reports are first reproduced as a working attack example and then privately passed to developers. This is a standard practice of responsible disclosure, but in this case, it raises questions about coordination given the scale of the findings.

Reason: attacks on Coldcard and the loss of $100 million

The campaign started amid the exploitation of a critical bug in Coldcard hardware wallets. In my assessment, this is one of the most serious incidents of the year: at least 15 different attackers took advantage of the vulnerability, and losses from three confirmed waves of attacks exceeded $100 million. Considering a suspected fourth wave, the amount could rise to $130 million. Manufacturer Coinkite explained that the problem arose after a firmware change in 2021: in some cases, devices used a predictable MicroPython software random number generator instead of the hardware one. The effective entropy of seed phrases dropped from 128 bits to approximately 40 bits on Mk2 and Mk3 models, and to 72 bits on Mk4, Mk5, and Q. Fixed firmware has already been released, but it does not protect already created seed phrases—owners are recommended to generate new ones on an updated device.

AI accelerates discovery, but not resolution

Hamilton noted that the main limitation of the campaign is not the discovery of issues but delivering reports to the right developers. So far, the team has made more than a dozen disclosures, scanning 150 repositories. Infrastructure costs amounted to about $20,000, and funding is already secured. This is an important signal: even with AI accelerating vulnerability discovery, human coordination remains the bottleneck. The ecosystem is currently in chaos—developers are literally flooded with reports, creating a risk that critical findings may be lost in the flow of information.

My comment: this incident highlights a systemic problem—the speed of automated vulnerability discovery outpaces the ecosystem's ability to process it. AI tools are becoming a powerful weapon for both researchers and attackers, and without standardized disclosure protocols, we risk facing a wave of exploits that will exceed the capabilities of responsible teams.