From my (quick) search, it seems like there's no readily available license (or some sort of standard clause) to prevent an openly accessible codebase to be used for AI training.<p>Does anyone know of any existing repository out there that states this in their license? Any advice on how to do it?<p>P.S.: I’m specifically looking for how to include this restriction in a license, I kindly ask stay on topic here.<p>Thanks in advance!
There's nothing stopping you from <i>putting</i> a no-AI-training provision in your license. However, it's functionally redundant with some licenses which implicitly disallow AI training anyways and would likely not be obeyed.<p>Whether it would give you legal standing to successfully sue anyone that did it anyways is still up for legal debate.
AI companies use the argument "what we're doing is fair use, so we don't need to follow the license". If that argument is valid, then you can't because nothing you put in the license will apply to them anyway. If that argument is invalid, then basically every license* already does opt out, since basically every license* requires attribution, and the AI companies don't give that.<p>* Even the extremely lax MIT license requires attribution. Basically the only licenses that don't are public-domain equivalent ones like CC0, the Unlicense, and the WTFPL: <a href="https://en.wikipedia.org/wiki/Public-domain-equivalent_license" rel="nofollow">https://en.wikipedia.org/wiki/Public-domain-equivalent_licen...</a>