What the new rules demand<\/h2>\n\n\n\n
The announcement on r\/opensource<\/a> goes far beyond AI. A project has to be at least a year old and actively maintained, have multiple contributors, and point to an existing community. School projects and software built for AI, such as MCP servers, are out. Anyone who posts needs a Reddit account that is at least a year old and has significant activity. And if you don’t speak English, you are told to find a different community.<\/p>\n\n\n\n The core fits into one sentence: \u201cIf it’s obvious you’re using AI in your project, we don’t want it.\u201d It is followed by the side swipe that gave the post its title: \u201cTokens are cheap, and we don’t care how many you spent to make your amazing project.\u201d Anyone who posts AI content or begs for an exception via modmail is banned for good. Only officially incorporated nonprofit organizations are exempt from the minimum requirements.<\/p>\n\n\n\n On r\/fossdroid, the language rule has since been struck through. Instead, the post now says it’s not about language or origin but about AI alone. Classic machine translation is allowed, translation via a language model is not. Whether DeepL or Google Translate still count as classic is anyone’s guess. Both have relied on neural networks for years, and Reddit now translates posts on its own.<\/p>\n\n\n\n How the rules came about is described on the blog<\/a> of Dan Brown, the developer of BookStack and, until recently, one of the moderators of r\/opensource. According to Brown, the rules came from Wolvereness alone, with nothing agreed with the rest of the team. Brown considers the AI rule unenforceable and the language rule hurtful. Brown has since stepped back as a moderator.<\/p>\n\n\n\n The timing is delicate. The Linux kernel is currently going the opposite way. At the Maintainers Summit in Prague on October 8, a majority was in favor of dropping<\/a> (in German) the \u201cAssisted-by\u201d tag for AI-assisted patches again. Only about ten percent of the 18,533 commits for Linux 7.3 up to -rc7 carry it, yet according to Sasha Levin, AI is used far more often. Developers are supposed to describe their use of AI in the cover letter from now on.<\/p>\n\n\n\n Linus Torvalds himself considers rules along the lines of \u201cdon’t use AI for X\u201d idiotic and instead expects developers to understand what they are doing. On October 13, he submitted a series of ten patches that, according to it-daily<\/a> (in German), he developed with Claude Code. Reviewers promptly found a race condition in them. So the kernel developers use AI quite openly, just with review.<\/p>\n\n\n\n No wonder \u201cWow r\/opensource banning Linux\u201d is among the top-voted comments under the announcement. Asked in all seriousness how the kernel would fare, Wolvereness replied that it would probably be allowed \u201con the subjective line.\u201d In other words, AI is banned, unless the mods decide to look the other way. Wolvereness admits elsewhere that there is no objective standard.<\/p>\n\n\n\n The backlash is fierce. The announcement sits at zero points, the top comment at more than 1,400: \u201cThis is orthogonal to open source.\u201d Whether software is free is decided by its license, not by the tool. Another user has maintained a package with more than 300 stars on GitHub for nine years, entirely without AI. It gets thrown out anyway: only one maintainer.<\/p>\n\n\n\n The language rule draws at least as much anger, and many simply call it xenophobic. A user from Japan deliberately writes by hand, in shaky English, that translation services are needed for reading too, just to understand everyone else. And critics read the exemption for incorporated organizations as meaning that large projects may use AI, while individual developers may not.<\/p>\n\n\n\n There is support, too. Many welcome the end of the endless \u201cI got tired of X so I built Y\u201d posts, and some are glad that someone is finally cracking down. One commenter on r\/fossdroid argues that prompted-together apps will never get the care developers give their own projects. Others counter that plenty of hand-written FOSS projects quietly die as well.<\/p>\n\n\n\n I can understand both sides. On the one hand, using language models in software development offers huge opportunities. It is simply incredibly fast, and yes, the code that comes out is often very usable. By the way, \u201cTokens are cheap\u201d is only partly true: a retired software developer in the thread does the math and pays 200 dollars a month in AI subscriptions for an open source project.<\/p>\n\n\n\n Especially for projects where volunteer developers aren’t exactly lining up, this is a huge opportunity. On r\/fossdroid, someone cites their own app ShizuCallRecorder as an example: 2,000 stars on GitHub, more than 103,000 downloads, but only two pull requests with code from outside, both created with AI help. How is a project like that supposed to reach three contributors? A blanket ban hits exactly these people.<\/p>\n\n\n\n What AI can do for security is shown by two graphs from Greg Kroah-Hartman’s repository on kernel development<\/a>. Up to Linux 6.19, each release consistently collected about 500 fixes for CVE vulnerabilities. With 7.0, the number jumps to more than 1,200, and with 7.2 to about 2,300. It’s hardly a coincidence that this happens right when AI tools for development and code analysis are picking up speed.<\/p>\n\n\n\nLinus Torvalds sees it very differently<\/h2>\n\n\n\n
What the community says<\/h2>\n\n\n\n
The opportunity: speed for projects without new blood<\/h2>\n\n\n\n
More holes found, fewer new ones built in<\/h2>\n\n\n\n
![\"Line](\"https:\/\/linuxundich.de\/wp-content\/uploads\/2026\/10\/kernel-cve-fixed-1280x960.webp\")