A volunteer team of sixteen Bitcoin developers has flagged 85 critical security vulnerabilities across 390 Bitcoin-related projects in just over 24 hours, using AI models to scan codebases at a pace no human audit could match.
The coordinated audit, led by developers including the pseudonymous Calle — creator of the Cashu ecash protocol — filed 4,962 total findings, including 635 high-severity issues in addition to the 85 critical ones. The targets included Bitcoin wallets, cryptographic libraries, and core infrastructure projects.
"Situation is extremely bad," Calle wrote on social media, describing the volume of critical findings.
AI as a Double-Edged Sword
The audit was conducted by pointing AI models — both commercial and open-source — at Bitcoin-related codebases, using a combination of manual oversight and increasingly automated harnesses built by developer Rob Hamilton. The group averaged roughly one critical bug per hour per person, at approximately $10,000 per day in compute costs.
The findings highlight how AI is transforming security research for both defenders and attackers. Most critical reports were quickly verified by project owners and reproduced using working proof-of-concept code in local test environments before being submitted.
But the volume is creating its own crisis. Calle acknowledged "a lot of chaos right now in the ecosystem" and apologized to maintainers overwhelmed by the flood of reports. The group is still refining its process to filter out false positives — what Calle called "the slop."
"The hardest part is coordinating to get things to the right people," Hamilton wrote. "While it is powerful, having found critical issues, I would view this as only version one."
The Attacker's Advantage
The audit lands in an ecosystem already reeling from a major exploit. The Coldcard hardware wallet hack, which began July 30, has drained as much as $114 million from wallets whose seed phrases were generated by faulty firmware. That bug had been dormant since 2021 and required no physical access to the device once the vulnerable key space was identified.
Attackers have access to the same AI tools. Anthropic disclosed in April that one of its models found a previously undiscovered bug in widely used encryption software that had been present for 27 years, at a cost of less than $50. Google's threat intelligence team separately reported in May that a criminal group had been preparing an attack built on a flaw identified by an AI model.
The implication is clear: the AI security arms race is no longer theoretical. Projects that do not proactively audit their code using these tools are racing against adversaries who already are.
The Bitcoin development community is now grappling with how to scale its response. The volunteer audit group is building better automated tooling, but the immediate challenge is operational — routing findings to the right maintainers and ensuring fixes are deployed before exploits are weaponized.