A volunteer Bitcoin security initiative has used frontier AI models to scan roughly 150 repositories and make more than a dozen disclosures, demonstrating how artificial intelligence could dramatically accelerate open-source security reviews. Artificial intelligence is rapidly becoming a security tool for Bitcoin developers, with a volunteer group using Moonshot AI’s Kimi K3 to search open-source Bitcoin software for vulnerabilities at a pace that would be difficult to reproduce through conventional manual auditing. The initiative, known as Bitcoin Red Team, emerged following security concerns surrounding Bitcoin wallet infrastructure. By early August, the group said it had scanned approximately 150 repositories and made more than a dozen responsible vulnerability disclosures.
Kimi K3 has been used for much of the intensive code analysis, with other frontier models supporting verification and documentation. The group has also experimented with additional AI security systems for particularly important parts of the Bitcoin ecosystem. The findings do not mean Bitcoin itself has been compromised. The repositories being examined span the wider Bitcoin software ecosystem, which includes wallets, libraries and infrastructure projects rather than simply Bitcoin Core.
Kimi K3 Becomes an AI Security Auditor
Bitcoin Red Team’s approach effectively turns frontier AI into an automated security researcher. The group has developed an agent harness designed to let Kimi K3 perform much of the computationally intensive vulnerability hunting, while models including GPT and Claude systems can assist with supporting documentation and review. The project had already spent roughly $20,000 on AI services by early August, according to public updates from the initiative, illustrating both the computational cost and potential scale of AI-driven code auditing.
Kimi K3 is particularly notable because Moonshot AI released the model as open weight. The 2.8-trillion-parameter mixture-of-experts system activates 104 billion parameters and supports a one-million-token context window, allowing it to process unusually large quantities of source code and documentation. Moonshot says K3 was specifically trained for long-horizon coding and agentic tasks, capabilities that make large software repositories a natural target for automated security analysis. However, AI-generated vulnerability reports still require human verification. Automated systems can produce false positives, misunderstand intended behavior or assign excessive severity to bugs that are difficult or impossible to exploit.
AI Changes the Economics of Open-Source Security
The larger implication for Bitcoin is the dramatic reduction in the cost and time required to inspect public code. Traditional security audits depend on scarce specialist engineers manually examining software. AI agents can instead evaluate repositories continuously, generate attack hypotheses and flag suspicious code for human researchers to reproduce. That creates an important defensive opportunity but also an obvious risk.
Bitcoin-related software is overwhelmingly open source. The same publicly accessible code that allows Bitcoin Red Team to identify vulnerabilities before they are exploited is also available to attackers equipped with increasingly capable AI systems. Kimi K3 itself illustrates the accelerating capabilities involved. Moonshot released its weights in July, making frontier-level coding technology available for independent deployment rather than restricting access through a centrally controlled service. Bitcoin Red Team ultimately plans to open-source its auditing harness so organizations can apply similar techniques to internal repositories.
That could mark an important change in cryptocurrency security. Open-source projects have historically relied on developers and independent researchers eventually discovering vulnerabilities. AI makes systematic inspection of hundreds of repositories economically feasible. The result is effectively an arms race: AI can give Bitcoin developers a much larger security-review workforce, but attackers increasingly have access to the same technology. The advantage may therefore belong to whichever side audits the code first.