Notes

• AI in ZK circuit auditing shows potential for identifying known vulnerabilities.
• SnarkSentinel performs well in comparing specifications and implementations.
• AI audit tools face challenges in identifying low-level encryption vulnerabilities.
• The codebase using Circom language is SnarkSentinel's first target.
• The team adopted RAG and proxy technologies to improve AI audit efficiency.
• Future audits may increasingly rely on human-AI collaboration.

• AI在审计ZK电路中的应用展示了识别已知漏洞的潜力。
• SnarkSentinel工具在比较规范与实现方面表现出色。
• AI审计工具面临识别低级别加密漏洞的挑战。
• 使用Circom语言的代码库是SnarkSentinel的首个目标。
• 团队采用了RAG和代理技术来提高AI的审计效率。
• 未来的审计可能会更加依赖人类与AI的合作。