A volunteer security initiative using advanced artificial intelligence has uncovered 4,962 software issues across 390 Bitcoin-related projects in a coordinated 30-hour campaign. The effort, led by pseudonymous developer Calle and supported by OpenSats, OpenCode, and AnchorWatch, combined human expertise with frontier AI models to identify weaknesses in wallets, cryptographic libraries, and infrastructure.
According to figures released by the team, the findings included 85 critical and 635 high-severity vulnerabilities, totaling 720 of the most serious reports. The review maintained a blistering pace of roughly 166 reported findings every hour, with crypto libraries and software development kits recording the largest share at 1,385 issues. Only one reviewed project completed the campaign without any reported problems.
AnchorWatch CEO Rob Hamilton disclosed that the group has spent about $20,000 on AI services while building the “Bitcoin red team” platform. The team utilized multiple cutting-edge models—Kimi K3, OpenAI’s GPT Sol, Anthropic’s Claude Fable and Opus, and Z.ai’s GLM 5.2—to generate vulnerability assessments and supporting documentation. Calle noted the initiative was averaging “one critical exploit per hour per person” and burning through $10,000 per day, underscoring both the intensity and high cost of the exercise.
Unlike traditional security audits, the campaign relied on human reviewers actively guiding AI systems, each using different prompts and methods to avoid blind spots. Verified critical findings have already been sent to affected project maintainers with proof-of-concept retest demonstrations, and many confirmations arrived quickly—though processing the sheer volume remains a significant challenge. The effort comes amid a broader industry trend where AI is increasingly used to find security flaws, such as the previously discovered Zcash counterfeiting vulnerability and recent Coldcard wallet exploit.
