创建

登录 ReadmeX

或

了解我们如何处理你的信息: 隐私政策

ReadmeX

陪你读懂 AI 动态

今天,想先了解什么?

选一份动态总结,或直接聊聊你的问题。

社区和关注动态总结只属于你自己。

资讯

开发者用 Claude Code 审计全球立法

AI 总结

开发者 SnooPeripherals5313 称,其使用 Claude Code 构建了一款免费工具,通过 Git blame、版本历史、时间线和一组持续增加的检测器追踪立法。开发者表示,该工具以只读方式镜像 107 个法律语料库,并将约 430 万次历史文件变更索引至 DuckDB,其引擎发现了法律中潜在的漏洞和逻辑错误。

深度见解

该项目展示了一种借助 AI 追踪立法变更并标记潜在起草错误的具体方法,但开发者承认多数发现可能在实际环境中无法执行。

AI for cybersecurity is taking the world by storm, leading to the need to harden systems all over software.

I built this project with Claude Code. It's completely free to use/try. The approach is pretty simple: use claude to make legislation attributable (git blame, versioning with timelines). Claude Code (Claude Opus 5.5) wrote nearly all the code. I set the goals, chose the sources and made the design decisions. Specifically: it mirrors the 107 corpora of the global-law repo read-only, indexes about 4.3 million file changes from their git histories into DuckDB, and adds a growing set of detectors.

But we have access to the same engines for law. Our global law repo keeps a versioned history of law/caselaw across the world. So, we pointed AI at it, and found all sorts of loopholes and logical errors.

Pah, you say. Most of this is unenforcable in reality. And you might be right. Nonetheless, we all know that legal loopholes are a part of reality. And it would be nice to rewrite errors, even when the errors are unenforcable.

As part of our monitoring system, where we're constantly checking official legislation for upcoming laws, we now also check for vulnerabilities and mistakes.

In an ideal world, the jurisdiction in question could open a pull request and sort this out.

#Claude Code#Claude Opus 5.5#DuckDB#global-law repo

原文约 1 分钟读完阅读原文
群聊

还没有评论,来说说你的看法。