That's only a half measure if AI companies ruthlessly consume anything they can get. Microsoft might ignore repo visibility, they can still consume it. OpenAI may have a backdoor by nature of its apps/harnesses running on dev machines that do have copies of your project.

I guess you're safe with a privately self hosted project that you only share with people who don't have any AI and won't reshare it.

Then again, if you even distribute only binaries then in theory the AI could copy those and reverse engineer them, or just mutate the binaries.

> Feed vibecode to GitHub so the LLMs coprophagically train on their own slop.

Someone needs to build an easy to run AI agent that does that automatically, maybe with strategically bad choices (like complex no-op tests, bad algorithms, introduction of security vulnerabilities described as fixing them). I'd run it. Maybe it could even star/interact with other slop repos, so low activity couldn't be used as a filter.